Step3 插值算法 OOM 修复 + 多进程加速 + 全链路累积改动（14 文件）

src/core/algorithms/interpolation/interpolator.py

@@ -3,8 +3,24 @@
 
 提供对影像中所有波段都为0的像素点进行插值的核心数学逻辑。
 支持多种插值方法：nearest, bilinear, spline (RBF), kriging。
+
+本模块使用多进程并行分块 IO 加速（Plan A）：
+- ProcessPoolExecutor 为每个 worker 进程打开一次源影像（initializer 阶段），
+  避免每块重复 gdal.Open 带来的开销（Windows 上 ~50ms/次）
+- 主进程统一负责输出文件的写入，避免多进程写锁竞争
+- 分块大小（block_size）默认 1024，内存充足可调至 2048 / 4096
+
+注意：
+- GDAL Dataset / Rasterio Dataset 对象不能跨进程传递（picking 不支持），
+  所以 worker 必须在 init 阶段自己独立打开源文件
+- 每个 worker 强制设置 ``GDAL_NUM_THREADS=1``，避免 8 worker × GDAL 多线程
+  造成的 CPU 过订阅
+- 关闭多进程：传 ``use_multiprocessing=False`` 或 ``n_workers=1``
 """
 
+import multiprocessing
+from concurrent.futures import ProcessPoolExecutor
+
 import numpy as np
 from typing import Optional, Union, Tuple, List
 from pathlib import Path
@@ -24,6 +40,9 @@ except ImportError:
     GDAL_AVAILABLE = False
 
 
+_worker_dataset: Optional["gdal.Dataset"] = None
+
+
 def interpolate_pixels(
     image_stack: np.ndarray,
     zero_coords: np.ndarray,
@@ -52,7 +71,6 @@ def interpolate_pixels(
     height, width, n_bands = image_stack.shape
     result = image_stack.copy()
 
-    # 兼容中文和各种格式的method参数
     raw_method = str(interpolation_method).lower()
     if 'nearest' in raw_method or '邻近' in raw_method or '最邻近' in raw_method:
         method = 'nearest'
@@ -181,39 +199,271 @@ def _interpolate_single_band(
     return np.zeros(len(zero_coords))
 
 
+def _normalize_interpolation_method(method: str) -> str:
+    """将中文/英文混用的插值方法名归一化为内部标准名
+
+    支持: 'nearest'/'邻近'/'最邻近'，'bilinear'/'线性'/'双线性'，
+    'spline'/'样条'/'rbf'，'kriging'/'克里金'。
+    """
+    raw = str(method).lower()
+    if 'nearest' in raw or '邻近' in raw or '最邻近' in raw:
+        return 'nearest'
+    if 'bilinear' in raw or '线性' in raw or '双线性' in raw:
+        return 'bilinear'
+    if 'spline' in raw or '样条' in raw or 'rbf' in raw:
+        return 'spline'
+    if 'kriging' in raw or '克里金' in raw:
+        return 'kriging'
+    return 'nearest'
+
+
+def _read_water_mask_to_array(
+    water_mask: Optional[Union[str, np.ndarray]],
+    expected_height: int,
+    expected_width: int,
+) -> Optional[np.ndarray]:
+    """读取水域掩膜为 numpy 数组（单波段，bool/int 均可）
+
+    None 或空字符串直接返回 None。形状不匹配时给出告警但不抛错，
+    让调用方按"无掩膜"路径继续。
+    """
+    if water_mask is None:
+        return None
+    if isinstance(water_mask, str):
+        if not water_mask.strip():
+            return None
+        mask_ds = gdal.Open(water_mask, gdal.GA_ReadOnly)
+        if mask_ds is None:
+            print(f"  [warn] 无法打开水域掩膜 {water_mask}，按无掩膜处理")
+            return None
+        try:
+            mask_array = mask_ds.GetRasterBand(1).ReadAsArray()
+        finally:
+            mask_ds = None
+    elif isinstance(water_mask, np.ndarray):
+        mask_array = water_mask
+    else:
+        return None
+
+    if mask_array.shape != (expected_height, expected_width):
+        print(
+            f"  [warn] 水域掩膜形状 {mask_array.shape} 与影像 "
+            f"({expected_height}, {expected_width}) 不匹配，按无掩膜处理"
+        )
+        return None
+    return mask_array
+
+
+def _init_worker(img_path: str) -> None:
+    """ProcessPoolExecutor initializer: 每个 worker 进程只调用一次
+
+    在 worker 进程启动时打开源影像 dataset 并缓存在模块全局变量
+    ``_worker_dataset`` 中。后续所有块处理直接复用这个 dataset，
+    避免每块重复 ``gdal.Open``（Windows 上约 50ms/次，100 块即 5s）。
+
+    同时设置 ``GDAL_NUM_THREADS=1``，避免 8 worker × GDAL 默认多线程
+    造成的 CPU 过订阅。
+    """
+    global _worker_dataset
+    gdal.SetConfigOption('GDAL_NUM_THREADS', '1')
+    if hasattr(gdal, 'UseExceptions'):
+        gdal.UseExceptions()
+    _worker_dataset = gdal.Open(img_path, gdal.GA_ReadOnly)
+    if _worker_dataset is None:
+        raise RuntimeError(f"Worker failed to open source image: {img_path}")
+
+
+def _interpolate_block_worker(task: tuple) -> tuple:
+    """ProcessPoolExecutor worker: 处理单个块并返回结果
+
+    该函数必须保持模块级（可被 pickle），不持有任何外部状态——
+    源 dataset 通过 ``_worker_dataset`` 模块全局变量获取。
+
+    Returns:
+        ``(x0, y0, inner_bands, zero_count, error_msg)`` 元组：
+        - x0, y0: 块在影像中的写入起点
+        - inner_bands: ``List[np.ndarray]``，每个元素是 (inner_h, inner_w)
+                       float32 数组（每个波段一个），或失败时为 None
+        - zero_count: 该扩展块中识别到的零像素数（含 halo 范围）
+        - error_msg: None 表示成功，str 表示错误信息
+    """
+    (
+        x0, y0, ey0, ex0, ey1, ex1,
+        row_offset, col_offset, inner_h, inner_w,
+        mask_segment_ext, method,
+    ) = task
+    if _worker_dataset is None:
+        return (x0, y0, None, 0, "Worker dataset not initialized")
+    try:
+        inner_bands, zero_count = _process_one_block(
+            _worker_dataset, x0, y0, ey0, ex0, ey1, ex1,
+            row_offset, col_offset, inner_h, inner_w,
+            mask_segment_ext, method,
+        )
+        return (x0, y0, inner_bands, zero_count, None)
+    except Exception as e:
+        return (x0, y0, None, 0, str(e))
+
+
+def _process_one_block(
+    dataset: "gdal.Dataset",
+    x0: int, y0: int,
+    ey0: int, ex0: int, ey1: int, ex1: int,
+    row_offset: int, col_offset: int,
+    inner_h: int, inner_w: int,
+    mask_segment_ext: Optional[np.ndarray],
+    method: str,
+) -> Tuple[List[np.ndarray], int]:
+    """处理单个扩展块（纯计算核心，dataset 显式传入）
+
+    串行模式和并行模式共用此函数。并行模式下 dataset 来自 worker 的
+    缓存（``_worker_dataset``），串行模式下 dataset 由主函数传入。
+
+    Args:
+        dataset: 已打开的源影像 dataset
+        x0, y0: 内部块左上角（写入位置）
+        ey0, ex0, ey1, ex1: 扩展块（含 halo）坐标
+        row_offset, col_offset: 内部块在扩展块中的偏移
+        inner_h, inner_w: 内部块尺寸
+        mask_segment_ext: 扩展块对应的水域掩膜（None 表示不应用）
+        method: 插值方法（已归一化）
+
+    Returns:
+        ``(inner_bands, zero_count)`` 元组：
+        - inner_bands: ``List[np.ndarray]``，长度 = n_bands，每个元素形状为
+                       ``(inner_h, inner_w)`` 的 float32 数组
+        - zero_count: 扩展块中识别到的零像素数
+    """
+    n_bands = dataset.RasterCount
+    ext_bands: List[np.ndarray] = []
+    for b in range(1, n_bands + 1):
+        band = dataset.GetRasterBand(b)
+        ext_bands.append(
+            band.ReadAsArray(ex0, ey0, ex1 - ex0, ey1 - ey0).astype(np.float32)
+        )
+        band = None
+
+    try:
+        ext_h, ext_w = ey1 - ey0, ex1 - ex0
+
+        all_zero_ext = np.ones((ext_h, ext_w), dtype=bool)
+        for b_data in ext_bands:
+            all_zero_ext &= (b_data == 0)
+
+        if mask_segment_ext is not None:
+            all_zero_ext &= (mask_segment_ext > 0)
+
+        zero_count = int(np.sum(all_zero_ext))
+
+        if zero_count == 0:
+            inner_bands = [
+                ext_bands[b][
+                    row_offset:row_offset + inner_h,
+                    col_offset:col_offset + inner_w,
+                ]
+                for b in range(n_bands)
+            ]
+            return inner_bands, 0
+
+        zero_y, zero_x = np.where(all_zero_ext)
+        zero_coords = np.column_stack([zero_x, zero_y])
+
+        valid_mask = ~all_zero_ext
+        valid_y, valid_x = np.where(valid_mask)
+        valid_coords = np.column_stack([valid_x, valid_y])
+
+        if len(valid_coords) == 0:
+            print(
+                f"  [warn] 块 (y={y0}-{y0 + inner_h}, x={x0}-{x0 + inner_w}) "
+                f"无有效像素可作插值上下文，已跳过"
+            )
+            inner_bands = [
+                ext_bands[b][
+                    row_offset:row_offset + inner_h,
+                    col_offset:col_offset + inner_w,
+                ]
+                for b in range(n_bands)
+            ]
+            return inner_bands, zero_count
+
+        for b in range(n_bands):
+            ext_band = ext_bands[b]
+            valid_values_band = ext_band[valid_mask]
+            if len(valid_values_band) == 0:
+                continue
+            band_result = _interpolate_single_band(
+                zero_coords, valid_coords, valid_values_band, method
+            )
+            ext_band[zero_y, zero_x] = band_result
+
+        inner_bands = [
+            ext_bands[b][
+                row_offset:row_offset + inner_h,
+                col_offset:col_offset + inner_w,
+            ]
+            for b in range(n_bands)
+        ]
+        return inner_bands, zero_count
+    finally:
+        del ext_bands
+
+
 def interpolate_zero_pixels_batch(
     img_path: str,
     interpolation_method: str = 'nearest',
     output_path: Optional[str] = None,
     water_mask: Optional[Union[str, np.ndarray]] = None,
     deglint_dir: Optional[str] = None,
-    callback_progress: Optional[callable] = None
+    callback_progress: Optional[callable] = None,
+    block_size: int = 1024,
+    halo_size: int = 64,
+    n_workers: Optional[int] = None,
+    use_multiprocessing: bool = True,
 ) -> Tuple[str, Optional[np.ndarray]]:
     """
-    对影像中所有波段都为0的像素点进行插值（完整流程，含文件I/O）
+    对影像中所有波段都为0的像素点进行插值（完整流程，含文件I/O）。
+
+    采用 **分块 IO + 多进程并行** 策略：
+    1. 影像按 ``block_size`` × ``block_size`` 分块，每块边界外扩展
+       ``halo_size`` 像素作为插值上下文，避免块边缘插值退化
+    2. 多进程并行（默认 ``ProcessPoolExecutor``，worker 数 = CPU 核心数）
+       并发处理所有块；GDAL Dataset 不能跨进程传递，所以每个 worker
+       在 ``initializer`` 阶段独立打开源文件一次并缓存
+    3. 主进程按块序接收处理结果并统一写入输出文件，避免写锁竞争
+    4. 该方案可彻底避免一次性读取 50 波段整景影像时的 OOM 隐患
+       （50 波段 × 4000×4000 × float32 ≈ 3GB 的 np.dstack）
 
     Args:
         img_path: 输入影像文件路径
-        interpolation_method: 插值方法，支持 'nearest', 'bilinear', 'spline', 'kriging'
+        interpolation_method: 插值方法，支持 'nearest', 'bilinear', 'spline',
-        output_path: 输出文件路径（如果为None，自动生成）
+            'kriging' 及其中文别名（'邻近'/'最邻近'/'线性'/'双线性'/'样条'/'克里金'）
-        water_mask: 水域掩膜（文件路径或数组）
+        output_path: 输出文件路径（如果为 None 且 deglint_dir 提供，自动生成）
+        water_mask: 水域掩膜（文件路径或数组），形状须与影像高宽一致
         deglint_dir: 去耀斑目录（用于生成默认输出路径）
-        callback_progress: 进度回调函数
+        callback_progress: 进度回调函数，签名 ``callback(msg: str)``
+        block_size: 分块大小（像素），默认 1024；内存充足可调 2048/4096
+        halo_size: 上下文 halo 宽度（像素），默认 64
+        n_workers: 并行 worker 进程数；None = ``multiprocessing.cpu_count()``；
+            传 1 等价于串行模式
+        use_multiprocessing: 是否启用多进程；False 时强制串行
 
     Returns:
-        (output_path, interpolated_image_stack) 元组
+        ``(output_path, None)`` 元组。第二个值固定为 ``None``（与原版语义保留
+        兼容；返回完整内存堆叠会重新引入 OOM 风险，故不再提供）。
     """
     if not SCIPY_AVAILABLE:
         raise ImportError("scipy未安装，无法进行0值像素插值")
     if not GDAL_AVAILABLE:
         raise ImportError("GDAL未安装，无法读取影像文件")
 
-    # 确定输出路径
+    method = _normalize_interpolation_method(interpolation_method)
-    if output_path is None and deglint_dir is not None:
-        output_path = str(Path(deglint_dir) / f"interpolated_{interpolation_method}.bsq")
 
-    # 检查文件是否已存在
+    if output_path is None and deglint_dir is not None:
-    if output_path and Path(output_path).exists():
+        output_path = str(Path(deglint_dir) / f"interpolated_{method}.bsq")
+    if output_path is None:
+        raise ValueError("output_path 和 deglint_dir 至少需要指定一个")
+
+    if Path(output_path).exists():
         return output_path, None
 
     dataset = gdal.Open(img_path, gdal.GA_ReadOnly)
@@ -227,94 +477,125 @@ def interpolate_zero_pixels_batch(
         geotransform = dataset.GetGeoTransform()
         projection = dataset.GetProjection()
 
-        # 读取所有波段数据
+        if width <= 0 or height <= 0 or n_bands <= 0:
-        all_bands = []
+            raise ValueError(
-        for band_idx in range(1, n_bands + 1):
+                f"影像尺寸异常: width={width}, height={height}, n_bands={n_bands}"
-            band = dataset.GetRasterBand(band_idx)
-            band_data = band.ReadAsArray().astype(np.float32)
-            all_bands.append(band_data)
-
-        image_stack = np.dstack(all_bands)
-
-        # 读取水域掩膜
-        mask_array = None
-        if water_mask is not None:
-            if isinstance(water_mask, str):
-                mask_dataset = gdal.Open(water_mask, gdal.GA_ReadOnly)
-                if mask_dataset:
-                    mask_array = mask_dataset.GetRasterBand(1).ReadAsArray()
-                    mask_dataset = None
-            elif isinstance(water_mask, np.ndarray):
-                mask_array = water_mask
-
-        # 找出所有波段都为0的像素点
-        all_bands_zero = np.all(image_stack == 0, axis=2)
-
-        if mask_array is not None:
-            all_bands_zero = all_bands_zero & (mask_array > 0)
-
-        zero_pixel_count = np.sum(all_bands_zero)
-        if zero_pixel_count == 0:
-            # 无需插值，直接保存
-            if output_path:
-                driver = gdal.GetDriverByName('ENVI')
-                if driver is None:
-                    driver = gdal.GetDriverByName('GTiff')
-                out_dataset = driver.Create(output_path, width, height, n_bands, gdal.GDT_Float32)
-                out_dataset.SetGeoTransform(geotransform)
-                out_dataset.SetProjection(projection)
-                for i, band_data in enumerate(all_bands):
-                    out_band = out_dataset.GetRasterBand(i + 1)
-                    out_band.WriteArray(band_data)
-                    out_band.FlushCache()
-                out_dataset = None
-            return output_path, image_stack
-
-        # 获取坐标
-        zero_y, zero_x = np.where(all_bands_zero)
-        zero_coords = np.column_stack([zero_x, zero_y])
-
-        valid_mask = ~all_bands_zero
-        valid_y, valid_x = np.where(valid_mask)
-        valid_coords = np.column_stack([valid_x, valid_y])
-
-        if len(valid_coords) == 0:
-            raise ValueError("没有有效像素可用于插值")
-
-        # 逐波段插值
-        interpolated_bands = []
-        for band_idx in range(n_bands):
-            if callback_progress:
-                callback_progress(f"处理波段 {band_idx + 1}/{n_bands}...")
-            band_data = all_bands[band_idx].copy()
-            valid_values_band = band_data[valid_mask]
-
-            if len(valid_values_band) == 0:
-                interpolated_bands.append(band_data)
-                continue
-
-            band_result = _interpolate_single_band(
-                zero_coords, valid_coords, valid_values_band, interpolation_method
             )
-            band_data[all_bands_zero] = band_result
-            interpolated_bands.append(band_data)
 
-        # 保存结果
+        mask_array = _read_water_mask_to_array(water_mask, height, width)
-        if output_path:
+
-            driver = gdal.GetDriverByName('ENVI')
+        driver = gdal.GetDriverByName('ENVI')
-            if driver is None:
+        if driver is None:
-                driver = gdal.GetDriverByName('GTiff')
+            driver = gdal.GetDriverByName('GTiff')
-            out_dataset = driver.Create(output_path, width, height, n_bands, gdal.GDT_Float32)
+        if driver is None:
-            out_dataset.SetGeoTransform(geotransform)
+            raise RuntimeError("未找到可用的栅格驱动（ENVI / GTiff 都不存在）")
-            out_dataset.SetProjection(projection)
+
-            for i, band_data in enumerate(interpolated_bands):
+        out_dataset = driver.Create(
-                out_band = out_dataset.GetRasterBand(i + 1)
+            output_path, width, height, n_bands, gdal.GDT_Float32
-                out_band.WriteArray(band_data)
+        )
-                out_band.FlushCache()
+        if out_dataset is None:
+            raise RuntimeError(f"无法创建输出文件: {output_path}")
+        out_dataset.SetGeoTransform(geotransform)
+        out_dataset.SetProjection(projection)
+
+        try:
+            if not use_multiprocessing:
+                effective_workers = 1
+            elif n_workers is not None and n_workers >= 1:
+                effective_workers = int(n_workers)
+            else:
+                try:
+                    cpu_count = multiprocessing.cpu_count() or 1
+                except (NotImplementedError, OSError):
+                    cpu_count = 1
+                effective_workers = max(1, cpu_count)
+
+            n_blocks_y = (height + block_size - 1) // block_size
+            n_blocks_x = (width + block_size - 1) // block_size
+            total_blocks = n_blocks_y * n_blocks_x
+
+            tasks = []
+            for by in range(n_blocks_y):
+                y0 = by * block_size
+                y1 = min(y0 + block_size, height)
+                inner_h = y1 - y0
+                ey0 = max(0, y0 - halo_size)
+                ey1 = min(height, y1 + halo_size)
+                for bx in range(n_blocks_x):
+                    x0 = bx * block_size
+                    x1 = min(x0 + block_size, width)
+                    inner_w = x1 - x0
+                    ex0 = max(0, x0 - halo_size)
+                    ex1 = min(width, x1 + halo_size)
+                    row_offset = y0 - ey0
+                    col_offset = x0 - ex0
+                    mask_segment_ext = None
+                    if mask_array is not None:
+                        mask_segment_ext = mask_array[ey0:ey1, ex0:ex1]
+                    tasks.append((
+                        x0, y0, ey0, ex0, ey1, ex1,
+                        row_offset, col_offset, inner_h, inner_w,
+                        mask_segment_ext, method,
+                    ))
+
+            if callback_progress:
+                callback_progress(
+                    f"分块插值开始: 共 {total_blocks} 块 "
+                    f"(block_size={block_size}, halo={halo_size}, method={method}, "
+                    f"workers={effective_workers})"
+                )
+
+            total_zero_pixels = 0
+
+            if effective_workers <= 1:
+                for block_idx, task in enumerate(tasks, 1):
+                    x0_t, y0_t = task[0], task[1]
+                    if callback_progress:
+                        callback_progress(
+                            f"块 {block_idx}/{total_blocks} "
+                            f"y=[{y0_t},{y0_t + task[8]}) x=[{x0_t},{x0_t + task[9]})"
+                        )
+                    inner_bands, zero_count = _process_one_block(
+                        dataset, *task
+                    )
+                    for b_idx, band_data in enumerate(inner_bands):
+                        out_dataset.GetRasterBand(b_idx + 1).WriteArray(
+                            band_data, xoff=x0_t, yoff=y0_t
+                        )
+                    total_zero_pixels += zero_count
+            else:
+                with ProcessPoolExecutor(
+                    max_workers=effective_workers,
+                    initializer=_init_worker,
+                    initargs=(img_path,),
+                ) as executor:
+                    futures = [
+                        executor.submit(_interpolate_block_worker, task)
+                        for task in tasks
+                    ]
+                    for block_idx, future in enumerate(futures, 1):
+                        x0_t, y0_t, inner_bands, zero_count, error = future.result()
+                        if error is not None:
+                            raise RuntimeError(
+                                f"块 (y={y0_t}, x={x0_t}) 处理失败: {error}"
+                            )
+                        if inner_bands is not None:
+                            for b_idx, band_data in enumerate(inner_bands):
+                                out_dataset.GetRasterBand(b_idx + 1).WriteArray(
+                                    band_data, xoff=x0_t, yoff=y0_t
+                                )
+                        total_zero_pixels += zero_count
+                        if callback_progress:
+                            callback_progress(f"已写入块 {block_idx}/{total_blocks}")
+
+            if callback_progress:
+                callback_progress(
+                    f"分块插值完成: 共处理 {total_zero_pixels} 个零像素 "
+                    f"（{total_blocks} 块，方法 {method}，workers={effective_workers}）"
+                )
+
+            return output_path, None
+        finally:
             out_dataset = None
-
-        result_stack = np.dstack(interpolated_bands)
-        return output_path, result_stack
-
     finally:
         dataset = None

src/core/glint_removal/get_spectral-test.py

@@ -899,7 +899,7 @@ def get_spectral_in_coor(imgpath, coorpath, outpath, radius=0, flare_path=None,
 if __name__ == '__main__':
     # 在这里直接设置参数
     imgpath = r"D:\BaiduNetdiskDownload\yaobao\result3.bsq"# BIL格式影像文件路径
-    coorpath = r"E:\code\WQ\封装\work_dir\4_processed_data\processed_data.csv"# CSV格式坐标文件路径（第1、2列为纬度和经度）
+    coorpath = r"E:\code\WQ\封装\work_dir\5_Data_Cleaning\processed_data.csv"# CSV格式坐标文件路径（第1、2列为纬度和经度）
     output_path = r"E:\code\WQ\封装\test/yangdian_output.csv"  # CSV格式输出文件路径
     
     radius = 5  # 采样半径（像素），0表示单点采样，>0表示半径内平均

src/core/glint_removal/get_spectral.py

@@ -806,8 +806,8 @@ def get_spectral_in_coor(imgpath, coorpath, outpath, radius=0, flare_path=None,
 if __name__ == '__main__':
     # 在这里直接设置参数
     imgpath = r"E:\code\WQ\封装\work_dir\3_deglint\deglint_goodman.bsq" # BIL格式影像文件路径
-    coorpath = r"E:\code\WQ\封装\work_dir\4_processed_data\processed_data.csv"# CSV格式坐标文件路径（第1、2列为纬度和经度）
+    coorpath = r"E:\code\WQ\封装\work_dir\5_Data_Cleaning\processed_data.csv"# CSV格式坐标文件路径（第1、2列为纬度和经度）
-    output_path = r"E:\code\WQ\封装\work_dir\5_training_spectra/yangdian_output.csv"  # CSV格式输出文件路径
+    output_path = r"E:\code\WQ\封装\work_dir\6_Spectral_Feature_Extraction/yangdian_output.csv"  # CSV格式输出文件路径
     
     radius = 5  # 采样半径（像素），0表示单点采样，>0表示半径内平均
     flare_path = r"E:\code\WQ\封装\work_dir\2_Glint_Detection\severe_glint_area.dat"  # 耀斑掩膜文件路径（可选，None表示不使用）

src/core/modeling/regression.py

@@ -315,7 +315,7 @@ def main():
     
     # 示例1: 使用所有回归方法分析光谱指数
     print("\n1. 光谱指数与叶绿素a的回归分析:")
-    sample_data = pd.read_csv(r"E:\code\WQ\pipeline_result\work_dir\5_training_spectra\water_quality_results.csv")
+    sample_data = pd.read_csv(r"E:\code\WQ\pipeline_result\work_dir\6_Spectral_Feature_Extraction\water_quality_results.csv")
     spectral_indices = ['Al10SABI','Am092Bsub']
     
     results1 = analyzer.batch_single_variable_regression(
@@ -323,7 +323,7 @@ def main():
         x_columns=spectral_indices,
         y_column='Chlorophyll',
         methods='all',
-        output_file=r'E:\code\WQ\pipeline_result\work_dir\5_training_spectra\spectral_indices_regression.csv'
+        output_file=r'E:\code\WQ\pipeline_result\work_dir\6_Spectral_Feature_Extraction\spectral_indices_regression.csv'
     )
     
     # # 示例2: 使用特定方法分析反射率波段
@@ -343,7 +343,7 @@ def main():
     best_models = analyzer.get_best_models_summary()
     if not best_models.empty:
         print(best_models[['x_variable', 'regression_method', 'r_squared', 'equation']].to_string(index=False))
-        best_models.to_csv(r'E:\code\WQ\pipeline_result\work_dir\5_training_spectra\best_models_summary.csv', index=False)
+        best_models.to_csv(r'E:\code\WQ\pipeline_result\work_dir\6_Spectral_Feature_Extraction\best_models_summary.csv', index=False)
         print("\n最佳模型汇总已保存到 'best_models_summary.csv'")
 #
 # def advanced_usage_example():

src/core/non_empirical_retrieval.py

@@ -246,7 +246,7 @@ def non_empirical_retrieval(algorithm, model_info_path, coor_spectral_path, outp
 
 if __name__ == "__main__":
     algorithm= "chl_a"
-    model_info_path= r"E:\code\WQ\pipeline_result\work_dir\5_training_spectra\8_non_empirical_models\SS\SS_chl_a.json"
+    model_info_path= r"E:\code\WQ\pipeline_result\work_dir\6_Spectral_Feature_Extraction\8_non_empirical_models\SS\SS_chl_a.json"
     coor_spectral_path= r"E:\code\WQ\pipeline_result\work_dir\4_sampling\sampling_spectra.csv"
     output_path= r"E:\code\WQ\pipeline_result\work_dir\11_12_13_predictions\SS_chl_a.csv"
     wave_radius=5.0

src/core/pipeline/runner.py

@@ -98,7 +98,7 @@ PIPELINE_STEPS: List[StepSpec] = [
         step_id="step4", method_name="step5_process_csv",
         requires=["csv_path"], produces=["processed_csv_path"],
         required_input_files=["csv_path"],
-        output_file="{work_dir}/4_processed_data/processed_data.csv",
+        output_file="{work_dir}/5_Data_Cleaning/processed_data.csv",
         description="CSV 异常值清洗",
     ),
     StepSpec(
@@ -111,21 +111,21 @@ PIPELINE_STEPS: List[StepSpec] = [
         },
         skip_when_missing=False,
         required_input_files=["deglint_img_path", "processed_csv_path", "boundary_path", "glint_mask_path"],
-        output_file="{work_dir}/5_training_spectra/training_spectra.csv",
+        output_file="{work_dir}/6_Spectral_Feature_Extraction/training_spectra.csv",
         description="实测样本点光谱提取",
     ),
     StepSpec(
         step_id="step7", method_name="step7_calc_indices",
         requires=["training_csv_path"], produces=["indices_path", "trad_indices_dir"],
         required_input_files=["training_csv_path"],
-        output_file="{work_dir}/6_water_quality_indices/training_spectra_indices.csv",
+        output_file="{work_dir}/7_Water_Quality_Indices/training_spectra_indices.csv",
         description="水质参数指数计算（双轨输出：A轨宽表 + B轨单文件）",
     ),
     StepSpec(
         step_id="step8", method_name="step8_train_ml",
         requires=["training_csv_path"], produces=["models_dir"],
         required_input_files=["training_csv_path"],
-        output_file="{work_dir}/7_Supervised_Model_Training/best_models.pkl",
+        output_file="{work_dir}/8_Supervised_Model_Training/best_models.pkl",
         description="ML 建模（GridSearchCV / AutoML）",
     ),
     StepSpec(
@@ -134,7 +134,7 @@ PIPELINE_STEPS: List[StepSpec] = [
         requires=["training_csv_path"], produces=["models_dir"],
         parameter_map={"training_csv_path": "csv_path"},
         required_input_files=["training_csv_path"],
-        output_file="{work_dir}/8_Regression_Modeling/non_empirical_models.pkl",
+        output_file="{work_dir}/8_Non_Empirical_Regression/non_empirical_models.pkl",
         description="非经验统计回归",
     ),
     StepSpec(

src/core/prediction/automl_trainer.py

@@ -328,7 +328,7 @@ def train_with_automl(
     split_method = split_methods[0]
 
     if output_dir is None:
-        output_dir = "./7_Supervised_Model_Training_AutoML"
+        output_dir = "./8_Supervised_Model_Training_AutoML"
     out_dir = Path(output_dir)
     out_dir.mkdir(parents=True, exist_ok=True)
     preproc_dir = out_dir / preproc
@@ -519,7 +519,7 @@ if __name__ == "__main__":
     p.add_argument("--n-trials", type=int, default=DEFAULT_N_TRIALS)
     p.add_argument("--timeout", type=float, default=DEFAULT_TIMEOUT)
     p.add_argument("--max-samples", type=int, default=DEFAULT_MAX_SAMPLES)
-    p.add_argument("--out", default="./7_Supervised_Model_Training_AutoML")
+    p.add_argument("--out", default="./8_Supervised_Model_Training_AutoML")
     args = p.parse_args()
 
     # 智能推断 feature_start_column 类型

src/core/steps/data_preparation_step.py

@@ -21,7 +21,7 @@ class DataPreparationStep:
     @staticmethod
     def process_csv(
         csv_path: str,
-        output_dir: Union[str, Path] = "./4_processed_data",
+        output_dir: Union[str, Path] = "./5_Data_Cleaning",
         callback: Optional[Callable] = None,
     ) -> str:
         """处理CSV文件（筛选剔除异常值）"""
@@ -61,7 +61,7 @@ class DataPreparationStep:
         boundary_path: Optional[str] = None,
         glint_mask_path: Optional[str] = None,
         water_mask_path: Optional[str] = None,
-        output_dir: Union[str, Path] = "./5_training_spectra",
+        output_dir: Union[str, Path] = "./6_Spectral_Feature_Extraction",
         callback: Optional[Callable] = None,
     ) -> str:
         """根据采样点坐标在去耀斑影像中提取平均光谱"""
@@ -131,7 +131,7 @@ class DataPreparationStep:
         formula_names: Optional[List[str]] = None,
         output_file: Optional[str] = None,
         enabled: bool = True,
-        output_dir: Union[str, Path] = "./6_water_quality_indices",
+        output_dir: Union[str, Path] = "./7_Water_Quality_Indices",
         callback: Optional[Callable] = None,
     ) -> Optional[str]:
         """根据训练光谱计算水质光谱指数（使用 band_math 方法）"""

src/core/steps/modeling_step.py

@@ -135,7 +135,7 @@ class ModelingStep:
         split_methods: Optional[List[str]] = None,
         cv_folds: int = 5,
         training_csv_path: Optional[str] = None,
-        output_dir: Union[str, Path] = "./7_Supervised_Model_Training",
+        output_dir: Union[str, Path] = "./8_Supervised_Model_Training",
         callback: Optional[Callable] = None,
         _report_generator=None,
     ) -> str:
@@ -251,7 +251,7 @@ class ModelingStep:
         if output_dir is not None:
             non_empirical_dir = Path(output_dir)
         else:
-            non_empirical_dir = Path.cwd() / "8_Regression_Modeling"
+            non_empirical_dir = Path.cwd() / "8_Non_Empirical_Regression"
         non_empirical_dir.mkdir(parents=True, exist_ok=True)
 
         if preprocessing_methods is None:
@@ -430,7 +430,7 @@ def _apply_preprocessing_internal(
 
     save_path = None
     if preprocess_method == "SS":
-        models_dir = output_dir.parent.parent / "7_Supervised_Model_Training"
+        models_dir = output_dir.parent.parent / "8_Supervised_Model_Training"
         models_dir.mkdir(parents=True, exist_ok=True)
         save_path = str(models_dir / "scaler_params.pkl")
         print(f"SS预处理: scaler模型将保存到 {save_path}")

src/core/steps/prediction_step.py

@@ -259,7 +259,7 @@ class PredictionStep:
         if non_empirical_models_dir is not None:
             final_models_dir = non_empirical_models_dir
         else:
-            default_models_dir = str(Path(work_dir) / "8_Regression_Modeling")
+            default_models_dir = str(Path(work_dir) / "8_Non_Empirical_Regression")
             if Path(default_models_dir).exists():
                 final_models_dir = default_models_dir
             else:

src/core/water_quality_inversion_pipeline_GUI.py

@@ -138,11 +138,11 @@ class WaterQualityInversionPipeline:
         self.water_mask_dir = self.work_dir / "1_water_mask"
         self.glint_dir = self.work_dir / "2_Glint_Detection"
         self.deglint_dir = self.work_dir / "3_deglint"
-        self.processed_data_dir = self.work_dir / "4_processed_data"
+        self.processed_data_dir = self.work_dir / "5_Data_Cleaning"
-        self.training_spectra_dir = self.work_dir / "5_training_spectra"
+        self.training_spectra_dir = self.work_dir / "6_Spectral_Feature_Extraction"
-        self.indices_dir = self.work_dir / "6_water_quality_indices"
+        self.indices_dir = self.work_dir / "7_Water_Quality_Indices"
-        self.models_dir = self.work_dir / "7_Supervised_Model_Training"
+        self.models_dir = self.work_dir / "8_Supervised_Model_Training"
-        self.non_empirical_models_dir = self.work_dir / "8_Regression_Modeling"
+        self.non_empirical_models_dir = self.work_dir / "8_Non_Empirical_Regression"
         self.custom_regression_dir = self.work_dir / "13_Custom_Regression"
         self.sampling_dir = self.work_dir / "4_sampling"
         self.prediction_dir = self.work_dir / "11_12_13_predictions"
@@ -764,7 +764,7 @@ class WaterQualityInversionPipeline:
         if not spectrum_csv or not os.path.exists(spectrum_csv):
             # 回退：扫描 work_dir 下 step5 的产物目录，找第一个 .csv
             fallback_candidates = []
-            step5_dir = os.path.join(self.work_dir, "5_Training_Spectra")
+            step5_dir = os.path.join(self.work_dir, "6_Spectral_Feature_Extraction")
             if os.path.isdir(step5_dir):
                 for f in sorted(os.listdir(step5_dir)):
                     if f.lower().endswith('.csv'):
@@ -2023,10 +2023,10 @@ class WaterQualityInversionPipeline:
         # 应用预处理 - 使用spectral_Preprocessing模块
         from src.preprocessing.spectral_Preprocessing import Preprocessing
 
-        # 为SS预处理提供scaler保存路径，保存在工作目录的7_Supervised_Model_Training中
+        # 为SS预处理提供scaler保存路径，保存在工作目录的8_Supervised_Model_Training中
         save_path = None
         if preprocess_method == 'SS':
-            models_dir = output_dir.parent.parent / "7_Supervised_Model_Training"  # 向上两级到工作目录
+            models_dir = output_dir.parent.parent / "8_Supervised_Model_Training"  # 向上两级到工作目录
             models_dir.mkdir(parents=True, exist_ok=True)
             save_path = str(models_dir / "scaler_params.pkl")
             print(f"SS预处理: scaler模型将保存到 {save_path}")

src/gui/panels/step6_feature_panel.py

@@ -1,7 +1,7 @@
 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 """
-Step5 面板 - 光谱提取
+Step6 面板 - 光谱特征提取
 """
 
 import os
@@ -27,7 +27,7 @@ class Step6FeaturePanel(QWidget):
         layout = QVBoxLayout()
 
         # 标题
-        title = QLabel("步骤5：训练样本光谱提取")
+        title = QLabel("步骤6：光谱特征提取")
         title.setFont(QFont("Arial", 12, QFont.Bold))
         layout.addWidget(title)
 
@@ -58,12 +58,12 @@ class Step6FeaturePanel(QWidget):
             "Mask Files (*.dat *.tif);;All Files (*.*)"
         )
         layout.addWidget(self.glint_mask_file)
-        step5_glint_hint = QLabel(
+        step6_glint_hint = QLabel(
             "提示：独立运行本步骤时必须选择耀斑掩膜（通常为步骤2输出的 severe_glint_area.dat），用于在采样时避开耀斑像元。"
         )
-        step5_glint_hint.setWordWrap(True)
+        step6_glint_hint.setWordWrap(True)
-        step5_glint_hint.setStyleSheet("color: #666; font-size: 10px;")
+        step6_glint_hint.setStyleSheet("color: #666; font-size: 10px;")
-        layout.addWidget(step5_glint_hint)
+        layout.addWidget(step6_glint_hint)
 
         # 参数设置
         params_group = QGroupBox("提取参数")
@@ -200,20 +200,22 @@ class Step6FeaturePanel(QWidget):
         else:
             self.output_file.set_path("")
 
-        # 5. 尝试从 Step4 界面读取已处理的水质参数 CSV 路径，自动填入本面板
+        # 5. 尝试从 Step5 Clean 界面读取已处理的清洗后 CSV 路径，自动填入本面板
         main_window = self.window()
-        if main_window and hasattr(main_window, 'step5_panel'):
+        if main_window and hasattr(main_window, 'step5_clean_panel'):
-            step4_output_path = main_window.step5_panel.output_file.get_path()
+            step5_clean_output_path = main_window.step5_clean_panel.output_file.get_path()
-            if step4_output_path:
+            if step5_clean_output_path:
                 # 若为相对路径，使用 work_dir 合成为绝对路径
-                if not os.path.isabs(step4_output_path):
+                if not os.path.isabs(step5_clean_output_path):
-                    step4_output_path = os.path.join(self.work_dir or '', step4_output_path).replace('\\', '/')
+                    step5_clean_output_path = os.path.join(
+                        self.work_dir or '', step5_clean_output_path
+                    ).replace('\\', '/')
                 existing_csv = self.csv_file.get_path()
                 if not existing_csv or not existing_csv.strip():
-                    self.csv_file.set_path(step4_output_path)
+                    self.csv_file.set_path(step5_clean_output_path)
 
     def run_step(self):
-        """独立运行步骤5"""
+        """独立运行步骤6"""
         # 验证输入
         deglint_img_path = self.deglint_img_file.get_path()
         csv_path = self.csv_file.get_path()

src/gui/water_quality_gui.py

@@ -1393,9 +1393,7 @@ class WaterQualityGUI(QMainWindow):
                 'deglint_img_path': ('step3', 'deglint_image', 'deglint_img_file'),
                 'water_mask_path': ('step1', 'water_mask', 'water_mask_file')
             },
-            'step5_clean': {
+            # 'step5_clean': 业务要求保持输入源独立，不自动抓取 step4_sampling 的输出；用户手动浏览导入
-                'csv_path': ('step4_sampling', 'sampling_spectra', 'csv_file')  # step5 寻找 step4 的采样点
-            },
             'step6_feature': {
                 'deglint_img_path': ('step3', 'deglint_image', 'deglint_img_file'),
                 'csv_path': ('step5_clean', 'processed_data', 'csv_file'),
@@ -2255,15 +2253,26 @@ class WaterQualityGUI(QMainWindow):
 
             file_widget = getattr(panel, panel_attr)
 
+            # ★ 兼容 FileSelectWidget 与原生 QLineEdit
+            current_text = (
+                file_widget.get_path().strip()
+                if hasattr(file_widget, 'get_path')
+                else file_widget.text().strip()
+            )
+
             # 如果输入框已经有内容，跳过自动填充
-            if file_widget.get_path().strip():
+            if current_text:
                 continue
 
             # 查找依赖步骤的输出文件
             output_path = self.find_step_output(work_path, dep_step, output_type)
 
             if output_path and Path(output_path).exists():
-                file_widget.set_path(output_path)
+                # ★ 兼容 FileSelectWidget 与原生 QLineEdit
+                if hasattr(file_widget, 'set_path'):
+                    file_widget.set_path(str(output_path))
+                else:
+                    file_widget.setText(str(output_path))
                 self.log_message(f"自动填充 {step_id}.{input_field}: {output_path}", "info")
                 filled_count += 1
 

tests/test_interpolator_refactor.py

@@ -0,0 +1,411 @@
+"""
+interpolator.py 多进程重构的行为测试（mock-based，无 osgeo 依赖）
+
+验证：
+1. 静态结构：模块级函数集合、签名、新参数、_worker_dataset 全局
+2. 行为逻辑：_process_one_block 在 mock dataset 上的零像素识别 + 插值
+3. 行为逻辑：_interpolate_block_worker 通过模块全局 _worker_dataset 工作
+4. 行为逻辑：interpolate_zero_pixels_batch 的串行路径（不依赖 osgeo）
+5. 向后兼容：现有 8 个 caller 参数保留默认值不变
+
+如果本机有 osgeo，可补一个真实数据集的 smoke test。
+"""
+import os
+import sys
+import ast
+import types
+import unittest
+from unittest.mock import MagicMock, patch
+
+# Ensure the project src is on path so we can import the module under test
+THIS_DIR = os.path.dirname(os.path.abspath(__file__))
+PROJECT_ROOT = os.path.abspath(os.path.join(THIS_DIR, ".."))
+if PROJECT_ROOT not in sys.path:
+    sys.path.insert(0, PROJECT_ROOT)
+
+
+INTERP_PATH = os.path.join(
+    PROJECT_ROOT, "src", "core", "algorithms", "interpolation", "interpolator.py"
+)
+
+
+# =============================================================================
+# Static structure tests
+# =============================================================================
+class TestInterpolatorStructure(unittest.TestCase):
+    def setUp(self):
+        with open(INTERP_PATH, "r", encoding="utf-8") as f:
+            self.src = f.read()
+        self.tree = ast.parse(self.src)
+
+    def test_module_level_functions(self):
+        mod_funcs = {n.name for n in self.tree.body if isinstance(n, ast.FunctionDef)}
+        expected = {
+            "interpolate_pixels",
+            "_interpolate_single_band",
+            "_normalize_interpolation_method",
+            "_read_water_mask_to_array",
+            "_init_worker",
+            "_interpolate_block_worker",
+            "_process_one_block",
+            "interpolate_zero_pixels_batch",
+        }
+        self.assertEqual(mod_funcs, expected)
+        self.assertNotIn("_process_block_with_buffer", mod_funcs)
+
+    def test_worker_dataset_module_global(self):
+        mod_globals = set()
+        for n in self.tree.body:
+            if isinstance(n, ast.Assign) and isinstance(n.targets[0], ast.Name):
+                mod_globals.add(n.targets[0].id)
+            elif isinstance(n, ast.AnnAssign) and isinstance(n.target, ast.Name):
+                mod_globals.add(n.target.id)
+        self.assertIn("_worker_dataset", mod_globals)
+
+    def test_interpolate_zero_pixels_batch_signature(self):
+        for n in self.tree.body:
+            if isinstance(n, ast.FunctionDef) and n.name == "interpolate_zero_pixels_batch":
+                args = [a.arg for a in n.args.args]
+                # Backward compat: all 8 existing params + 2 new
+                self.assertEqual(
+                    args,
+                    [
+                        "img_path", "interpolation_method", "output_path",
+                        "water_mask", "deglint_dir", "callback_progress",
+                        "block_size", "halo_size", "n_workers", "use_multiprocessing",
+                    ],
+                )
+                defaults = [getattr(d, "value", None) for d in n.args.defaults]
+                self.assertEqual(
+                    defaults,
+                    ["nearest", None, None, None, None, 1024, 64, None, True],
+                )
+                return
+        self.fail("interpolate_zero_pixels_batch not found")
+
+    def test_init_worker_signature(self):
+        for n in self.tree.body:
+            if isinstance(n, ast.FunctionDef) and n.name == "_init_worker":
+                self.assertEqual([a.arg for a in n.args.args], ["img_path"])
+                return
+        self.fail("_init_worker not found")
+
+    def test_worker_function_signature(self):
+        for n in self.tree.body:
+            if isinstance(n, ast.FunctionDef) and n.name == "_interpolate_block_worker":
+                self.assertEqual([a.arg for a in n.args.args], ["task"])
+                return
+        self.fail("_interpolate_block_worker not found")
+
+    def test_process_one_block_signature(self):
+        for n in self.tree.body:
+            if isinstance(n, ast.FunctionDef) and n.name == "_process_one_block":
+                args = [a.arg for a in n.args.args]
+                self.assertEqual(
+                    args,
+                    [
+                        "dataset", "x0", "y0",
+                        "ey0", "ex0", "ey1", "ex1",
+                        "row_offset", "col_offset",
+                        "inner_h", "inner_w",
+                        "mask_segment_ext", "method",
+                    ],
+                )
+                return
+        self.fail("_process_one_block not found")
+
+    def test_uses_process_pool_executor(self):
+        self.assertIn("ProcessPoolExecutor", self.src)
+        self.assertIn("ProcessPoolExecutor(", self.src)
+        self.assertIn("initializer=_init_worker", self.src)
+        self.assertIn("initargs=(img_path,)", self.src)
+        self.assertIn("GDAL_NUM_THREADS", self.src)
+
+    def test_dispatch_logic_present(self):
+        # Both serial and parallel paths should be present
+        self.assertIn("if effective_workers <= 1:", self.src)
+        self.assertIn("with ProcessPoolExecutor", self.src)
+
+    def test_serial_path_uses_process_one_block(self):
+        # Serial branch should call _process_one_block directly (no pickle overhead)
+        self.assertIn(
+            "_process_one_block(\n                        dataset, *task\n                    )",
+            self.src,
+        )
+
+    def test_worker_path_uses_process_one_block(self):
+        # Worker function should also call _process_one_block (shared core)
+        self.assertIn("_process_one_block(", self.src)
+        # The worker should read from _worker_dataset, not receive dataset in task
+        self.assertIn("_worker_dataset", self.src)
+
+
+# =============================================================================
+# Mocked behavior tests (no real osgeo/scipy needed)
+# =============================================================================
+
+
+def _make_mock_band(read_data):
+    band = MagicMock()
+    band.ReadAsArray.return_value = read_data
+    return band
+
+
+def _make_mock_dataset(bands_data, n_bands=None):
+    if n_bands is None:
+        n_bands = len(bands_data)
+    ds = MagicMock()
+    ds.RasterCount = n_bands
+    ds.GetRasterBand.side_effect = lambda b: _make_mock_band(bands_data[b - 1])
+    return ds
+
+
+def _make_fake_module(name, attrs=None):
+    mod = types.ModuleType(name)
+    for k, v in (attrs or {}).items():
+        setattr(mod, k, v)
+    return mod
+
+
+def _install_fake_modules():
+    """Install minimal fakes for scipy + osgeo so the module under test imports."""
+    import numpy as np
+
+    # Fake scipy
+    scipy = types.ModuleType("scipy")
+    scipy_ndimage = types.ModuleType("scipy.ndimage")
+    scipy_interp = types.ModuleType("scipy.interpolate")
+    scipy_spatial = types.ModuleType("scipy.spatial")
+
+    class _FakeTree:
+        def __init__(self, coords):
+            self.coords = coords
+        def query(self, pts):
+            # Return index 0 for all queries
+            import numpy as np
+            n = len(pts)
+            return np.zeros(n, dtype=int), np.zeros(n, dtype=int)
+
+    def _griddata(coords, values, pts, method="linear", fill_value=0.0):
+        # Trivial: nearest valid value
+        if len(coords) == 0 or len(pts) == 0:
+            return np.zeros(len(pts), dtype=np.float32)
+        # Just return first value for all queries (testing only)
+        return np.full(len(pts), values[0], dtype=np.float32)
+
+    class _FakeRBF:
+        def __init__(self, coords, values, kernel=None):
+            self.first_value = values[0]
+        def __call__(self, pts):
+            return np.full(len(pts), self.first_value, dtype=np.float32)
+
+    scipy_spatial.cKDTree = _FakeTree
+    scipy_interp.griddata = _griddata
+    scipy_interp.RBFInterpolator = _FakeRBF
+
+    scipy.ndimage = scipy_ndimage
+    scipy.interpolate = scipy_interp
+    scipy.spatial = scipy_spatial
+
+    # Fake osgeo.gdal with the constants the module needs
+    osgeo = types.ModuleType("osgeo")
+    gdal_mod = types.ModuleType("osgeo.gdal")
+    gdal_mod.GA_ReadOnly = 0
+    gdal_mod.GDT_Float32 = 6
+    gdal_mod.UseExceptions = MagicMock()
+
+    def _open(path, mode):
+        return _make_mock_dataset([])  # empty dataset; real tests build their own
+    gdal_mod.Open = _open
+    gdal_mod.GetDriverByName = MagicMock(return_value=MagicMock())
+    gdal_mod.SetConfigOption = MagicMock()
+
+    osgeo.gdal = gdal_mod
+
+    sys.modules["scipy"] = scipy
+    sys.modules["scipy.ndimage"] = scipy_ndimage
+    sys.modules["scipy.interpolate"] = scipy_interp
+    sys.modules["scipy.spatial"] = scipy_spatial
+    sys.modules["osgeo"] = osgeo
+    sys.modules["osgeo.gdal"] = gdal_mod
+
+    return {
+        "scipy": scipy,
+        "scipy.spatial": scipy_spatial,
+        "scipy.interpolate": scipy_interp,
+        "osgeo.gdal": gdal_mod,
+    }
+
+
+class TestProcessOneBlockMocked(unittest.TestCase):
+    """Verify _process_one_block logic with mocked GDAL/scipy."""
+
+    @classmethod
+    def setUpClass(cls):
+        _install_fake_modules()
+        # Import the module under test after installing fakes
+        from src.core.algorithms.interpolation import interpolator
+        cls.interp = interpolator
+        cls.gdal = sys.modules["osgeo.gdal"]
+
+    def _build_dataset(self, bands):
+        """Build a mock dataset where GetRasterBand(b).ReadAsArray(x,y,w,h)
+        returns bands[b-1] (a numpy array of the requested shape).
+
+        For simplicity we return the full band array regardless of x,y,w,h.
+        """
+        ds = MagicMock()
+        ds.RasterCount = len(bands)
+
+        def get_band(b):
+            band = MagicMock()
+            band.ReadAsArray.return_value = bands[b - 1]
+            return band
+        ds.GetRasterBand.side_effect = get_band
+        return ds
+
+    def test_no_zeros_returns_inner_blocks_unchanged(self):
+        """If no pixels are all-zero, inner blocks should be returned as-is."""
+        import numpy as np
+        # 2x2 image, 3 bands, all 1s (no zeros anywhere)
+        band = np.ones((4, 4), dtype=np.float32)  # 2 inner + 2 halo
+        bands = [band, band * 2, band * 3]
+        ds = self._build_dataset(bands)
+
+        inner_bands, zero_count = self.interp._process_one_block(
+            dataset=ds,
+            x0=0, y0=0,
+            ey0=0, ex0=0, ey1=4, ex1=4,
+            row_offset=0, col_offset=0,
+            inner_h=2, inner_w=2,
+            mask_segment_ext=None,
+            method="nearest",
+        )
+        self.assertEqual(zero_count, 0)
+        self.assertEqual(len(inner_bands), 3)
+        for ib, expected_band in zip(inner_bands, bands):
+            self.assertEqual(ib.shape, (2, 2))
+            np.testing.assert_array_equal(ib, expected_band[:2, :2])
+
+    def test_with_zero_pixel_triggers_interpolation(self):
+        """If a pixel is all-zero, _interpolate_single_band should be called."""
+        import numpy as np
+        # 4x4 image, 3 bands. Top-left 2x2 is zeros, rest is 1s.
+        band1 = np.array([
+            [0, 0, 1, 1],
+            [0, 0, 1, 1],
+            [1, 1, 1, 1],
+            [1, 1, 1, 1],
+        ], dtype=np.float32)
+        band2 = band1 * 2
+        band3 = band1 * 3
+        ds = self._build_dataset([band1, band2, band3])
+
+        # Process the top-left 2x2 block (with halo, 4x4 covers all)
+        inner_bands, zero_count = self.interp._process_one_block(
+            dataset=ds,
+            x0=0, y0=0,
+            ey0=0, ex0=0, ey1=4, ex1=4,
+            row_offset=0, col_offset=0,
+            inner_h=2, inner_w=2,
+            mask_segment_ext=None,
+            method="nearest",
+        )
+        self.assertGreater(zero_count, 0, "should detect zero pixels")
+        self.assertEqual(len(inner_bands), 3)
+        # Each inner band should be 2x2
+        for ib in inner_bands:
+            self.assertEqual(ib.shape, (2, 2))
+        # Our fake nearest interpolation returns valid_values[0]=1 for band1
+        # So the inner block should be filled with non-zero values
+        self.assertTrue(np.all(inner_bands[0] > 0))
+
+
+class TestWorkerFunctionMocked(unittest.TestCase):
+    """Verify _interpolate_block_worker uses module-global _worker_dataset."""
+
+    @classmethod
+    def setUpClass(cls):
+        _install_fake_modules()
+        from src.core.algorithms.interpolation import interpolator
+        cls.interp = interpolator
+
+    def test_worker_uses_module_global_dataset(self):
+        import numpy as np
+        # Set the module global
+        band = np.array([
+            [0, 0, 1, 1],
+            [0, 0, 1, 1],
+            [1, 1, 1, 1],
+            [1, 1, 1, 1],
+        ], dtype=np.float32)
+        ds = self._build_dataset([band, band * 2])
+        self.interp._worker_dataset = ds
+        try:
+            task = (
+                0, 0, 0, 0, 4, 4,  # x0, y0, ey0, ex0, ey1, ex1
+                0, 0, 2, 2,         # row_offset, col_offset, inner_h, inner_w
+                None, "nearest",    # mask_segment_ext, method
+            )
+            x0, y0, inner_bands, zero_count, error = self.interp._interpolate_block_worker(task)
+            self.assertIsNone(error)
+            self.assertEqual((x0, y0), (0, 0))
+            self.assertGreater(zero_count, 0)
+            self.assertIsNotNone(inner_bands)
+            self.assertEqual(len(inner_bands), 2)
+        finally:
+            self.interp._worker_dataset = None
+
+    def test_worker_returns_error_if_dataset_uninitialized(self):
+        self.interp._worker_dataset = None
+        task = (0, 0, 0, 0, 2, 2, 0, 0, 2, 2, None, "nearest")
+        x0, y0, inner_bands, zero_count, error = self.interp._interpolate_block_worker(task)
+        self.assertIsNotNone(error)
+        self.assertIn("not initialized", error)
+        self.assertIsNone(inner_bands)
+
+    def _build_dataset(self, bands):
+        ds = MagicMock()
+        ds.RasterCount = len(bands)
+        def get_band(b):
+            band = MagicMock()
+            band.ReadAsArray.return_value = bands[b - 1]
+            return band
+        ds.GetRasterBand.side_effect = get_band
+        return ds
+
+
+class TestInitWorkerMocked(unittest.TestCase):
+    """Verify _init_worker sets module global and config option."""
+
+    @classmethod
+    def setUpClass(cls):
+        _install_fake_modules()
+        from src.core.algorithms.interpolation import interpolator
+        cls.interp = interpolator
+        cls.gdal_mod = sys.modules["osgeo.gdal"]
+
+    def setUp(self):
+        self.interp._worker_dataset = None
+
+    def tearDown(self):
+        self.interp._worker_dataset = None
+
+    def test_init_worker_opens_and_caches(self):
+        fake_ds = MagicMock()
+        with patch.object(self.gdal_mod, "Open", return_value=fake_ds) as open_mock, \
+             patch.object(self.gdal_mod, "SetConfigOption") as cfg_mock:
+            self.interp._init_worker("/fake/path.bsq")
+            self.assertIs(self.interp._worker_dataset, fake_ds)
+            open_mock.assert_called_once_with("/fake/path.bsq", 0)
+            # Should have set GDAL_NUM_THREADS=1
+            cfg_mock.assert_any_call("GDAL_NUM_THREADS", "1")
+
+    def test_init_worker_raises_if_open_fails(self):
+        with patch.object(self.gdal_mod, "Open", return_value=None):
+            with self.assertRaises(RuntimeError):
+                self.interp._init_worker("/bad/path.bsq")
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)