fix: 补齐缺失的 handler 文件 + IDW 插值退化检测 + V1 代码归档
This commit is contained in:
duxin
@ -556,7 +556,14 @@ class ContentMapper:
|
||||
return grid_content
|
||||
|
||||
def _perform_interpolation(self, points, values, grid_xx, grid_yy):
|
||||
"""执行空间插值"""
|
||||
"""三级降级插值策略：Kriging → IDW → 最近邻
|
||||
|
||||
2026-07-01 重构：
|
||||
- Kriging 优先（自动拟合球形变异函数，不强制 nugget）
|
||||
- Kriging 退化检测：若结果标准差接近 0（纯色图），自动回退 IDW
|
||||
- IDW 作为首选回退：无需拟合变异函数，不会产生纯色图
|
||||
- scipy linear/nearest 作为最后兜底
|
||||
"""
|
||||
print(f"插值输入检查：")
|
||||
print(f" - 数据点数量: {len(points)}")
|
||||
print(f" - 数据值范围: {values.min():.4f} - {values.max():.4f}")
|
||||
@ -573,149 +580,145 @@ class ContentMapper:
|
||||
if len(points) < 3:
|
||||
raise ValueError(f"有效数据点不足3个（当前：{len(points)}个）")
|
||||
|
||||
# 优先使用Kriging插值
|
||||
# ── 策略 0：值域极窄时直接跳过 Kriging ──
|
||||
value_range = float(values.max() - values.min())
|
||||
value_std = float(np.std(values))
|
||||
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
# 策略 1：Kriging（自动拟合球形变异函数）
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
kriging_degraded = False
|
||||
if PYKRIGE_AVAILABLE:
|
||||
try:
|
||||
print("正在使用Kriging插值（半变异函数模型，块金值=100%）...")
|
||||
print("正在使用 Kriging 插值（球形模型，自动拟合 nugget）...")
|
||||
grid_x = grid_xx[0, :]
|
||||
grid_y = grid_yy[:, 0]
|
||||
ok = OrdinaryKriging(
|
||||
points[:, 0], points[:, 1], values,
|
||||
variogram_model='spherical',
|
||||
verbose=False,
|
||||
enable_plotting=False
|
||||
enable_plotting=False,
|
||||
)
|
||||
# ★ 局部邻域 Kriging：只参考最近的 15 个点，避免万阶矩阵求逆
|
||||
z, _ = ok.execute('grid', grid_x, grid_y, backend='loop', n_closest_points=15)
|
||||
z, ss = ok.execute('grid', grid_x, grid_y, backend='loop', n_closest_points=15)
|
||||
grid_content = np.array(z)
|
||||
valid_count = np.sum(~np.isnan(grid_content))
|
||||
print(f"Kriging插值成功，有效点数: {valid_count} / {grid_content.size}")
|
||||
if valid_count > 0:
|
||||
return grid_content
|
||||
else:
|
||||
print("警告：Kriging插值结果为空，将回退到其他插值方法")
|
||||
except Exception as e:
|
||||
print(f"Kriging插值失败: {e}，将回退到其他插值方法")
|
||||
else:
|
||||
print("警告：pykrige未安装，无法使用Kriging插值，将使用其他插值方法")
|
||||
valid_mask = ~np.isnan(grid_content)
|
||||
valid_count = int(np.sum(valid_mask))
|
||||
|
||||
if valid_count > 0:
|
||||
# ★ 退化检测：若插值结果标准差 < 原始数据标准差的 5%，判定为纯色图
|
||||
kriging_std = float(np.nanstd(grid_content))
|
||||
degradation_ratio = kriging_std / max(value_std, 1e-12)
|
||||
print(f"Kriging 完成: 有效点={valid_count}/{grid_content.size}, "
|
||||
f"输出std={kriging_std:.6f}, 退化比={degradation_ratio:.3f}")
|
||||
|
||||
if degradation_ratio < 0.05 and value_range > 1e-8:
|
||||
print(f"⚠ Kriging 严重退化（输出 std/输入 std={degradation_ratio:.3f}<5%），"
|
||||
f"判定为纯色图，回退 IDW")
|
||||
kriging_degraded = True
|
||||
else:
|
||||
print(f"Kriging 通过退化检测，直接使用")
|
||||
return grid_content
|
||||
else:
|
||||
print("Kriging 结果全为 NaN，回退")
|
||||
kriging_degraded = True
|
||||
except Exception as e:
|
||||
print(f"Kriging 失败: {e}")
|
||||
kriging_degraded = True
|
||||
else:
|
||||
print("pykrige 未安装，跳过 Kriging")
|
||||
kriging_degraded = True
|
||||
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
# 策略 2：IDW（反距离权重）— 不需要拟合变异函数，绝不纯色
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
if kriging_degraded:
|
||||
try:
|
||||
print("正在使用 IDW 插值（反距离权重, power=2, neighbors=15）...")
|
||||
grid_content = self._idw_interpolation(
|
||||
points, values, grid_xx, grid_yy,
|
||||
power=2, n_neighbors=min(15, len(points)),
|
||||
)
|
||||
valid_count = int(np.sum(~np.isnan(grid_content)))
|
||||
if valid_count > 0:
|
||||
idw_std = float(np.nanstd(grid_content))
|
||||
print(f"IDW 完成: 有效点={valid_count}/{grid_content.size}, 输出std={idw_std:.6f}")
|
||||
if idw_std > 0:
|
||||
return grid_content
|
||||
else:
|
||||
print("IDW std=0（所有输入值完全相同），结果可用")
|
||||
return grid_content
|
||||
else:
|
||||
print("IDW 结果全为 NaN，回退 scipy 插值")
|
||||
except Exception as e:
|
||||
print(f"IDW 失败: {e}，回退 scipy 插值")
|
||||
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
# 策略 3：scipy 线性插值 + 最近邻填充（最终兜底）
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
try:
|
||||
# 首先尝试使用线性插值
|
||||
print("正在尝试线性插值...")
|
||||
print("正在尝试 scipy 线性插值...")
|
||||
grid_content = griddata(
|
||||
points, values, (grid_xx, grid_yy),
|
||||
method='linear', fill_value=np.nan
|
||||
)
|
||||
|
||||
# 检查线性插值结果
|
||||
valid_linear = ~np.isnan(grid_content)
|
||||
valid_count = np.sum(valid_linear)
|
||||
print(f"线性插值结果：有效点数 {valid_count} / {grid_content.size}")
|
||||
valid_count = int(np.sum(~np.isnan(grid_content)))
|
||||
print(f"线性插值: 有效点={valid_count}/{grid_content.size}")
|
||||
|
||||
if valid_count > 0:
|
||||
print(f"线性插值成功，有效区域覆盖率: {valid_count / grid_content.size * 100:.1f}%")
|
||||
|
||||
# 如果有NaN值，用最近邻插值填充
|
||||
nan_count = np.sum(np.isnan(grid_content))
|
||||
nan_count = int(np.sum(np.isnan(grid_content)))
|
||||
if nan_count > 0:
|
||||
print(f"正在用最近邻插值填充 {nan_count} 个缺失值...")
|
||||
print(f"用最近邻填充 {nan_count} 个 NaN...")
|
||||
grid_nearest = griddata(
|
||||
points, values, (grid_xx, grid_yy),
|
||||
method='nearest'
|
||||
points, values, (grid_xx, grid_yy), method='nearest'
|
||||
)
|
||||
# 只填充线性插值的NaN区域
|
||||
nan_mask = np.isnan(grid_content)
|
||||
grid_content[nan_mask] = grid_nearest[nan_mask]
|
||||
print("缺失值填充完成")
|
||||
|
||||
# 最终检查
|
||||
final_valid = ~np.isnan(grid_content)
|
||||
print(f"最终有效点数: {np.sum(final_valid)} / {grid_content.size}")
|
||||
|
||||
grid_content[np.isnan(grid_content)] = grid_nearest[np.isnan(grid_content)]
|
||||
return grid_content
|
||||
else:
|
||||
print("线性插值失败，尝试最近邻插值...")
|
||||
|
||||
except Exception as e:
|
||||
print(f"线性插值失败: {e}")
|
||||
print("尝试最近邻插值...")
|
||||
|
||||
try:
|
||||
# 使用最近邻插值作为备选方案
|
||||
print("执行最近邻插值...")
|
||||
grid_content = griddata(
|
||||
points, values, (grid_xx, grid_yy),
|
||||
method='nearest'
|
||||
)
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
# 策略 4：最近邻（绝对兜底）
|
||||
# ═══════════════════════════════════════════════════════════
|
||||
print("执行最近邻插值（最终兜底）...")
|
||||
grid_content = griddata(
|
||||
points, values, (grid_xx, grid_yy), method='nearest'
|
||||
)
|
||||
if np.sum(~np.isnan(grid_content)) == 0:
|
||||
raise ValueError("所有插值方法均失败")
|
||||
return grid_content
|
||||
|
||||
valid_count = np.sum(~np.isnan(grid_content))
|
||||
print(f"最近邻插值成功，有效点数: {valid_count}")
|
||||
@staticmethod
|
||||
def _idw_interpolation(points, values, grid_xx, grid_yy,
|
||||
power=2, n_neighbors=15):
|
||||
"""IDW（反距离权重）插值 — 不需拟合模型，绝不产生纯色图。
|
||||
|
||||
if valid_count == 0:
|
||||
raise ValueError("最近邻插值也没有产生有效结果")
|
||||
Parameters:
|
||||
points: (N, 2) 采样点坐标
|
||||
values: (N,) 采样点值
|
||||
grid_xx, grid_yy: meshgrid 网格
|
||||
power: 距离衰减幂参数（默认 2）
|
||||
n_neighbors: 每个网格点参考的最近邻数量
|
||||
"""
|
||||
from scipy.spatial import cKDTree
|
||||
grid_shape = grid_xx.shape
|
||||
grid_flat = np.column_stack((grid_xx.ravel(), grid_yy.ravel()))
|
||||
values_flat = values.ravel()
|
||||
|
||||
return grid_content
|
||||
|
||||
except Exception as e:
|
||||
print(f"最近邻插值也失败: {e}")
|
||||
|
||||
# 对于地理坐标系，尝试更简单的方法
|
||||
if self.output_crs == 'EPSG:4326':
|
||||
print("地理坐标系检测到，尝试简化插值...")
|
||||
try:
|
||||
# 创建一个基于距离的简单插值
|
||||
grid_content = np.full(grid_xx.shape, np.nan)
|
||||
|
||||
# 为每个网格点找到最近的数据点
|
||||
for i in range(grid_xx.shape[0]):
|
||||
for j in range(grid_xx.shape[1]):
|
||||
grid_x, grid_y = grid_xx[i, j], grid_yy[i, j]
|
||||
|
||||
# 计算到所有数据点的距离
|
||||
distances = np.sqrt((points[:, 0] - grid_x) ** 2 + (points[:, 1] - grid_y) ** 2)
|
||||
nearest_idx = np.argmin(distances)
|
||||
|
||||
# 如果距离不是太远，就使用该值
|
||||
if distances[nearest_idx] < (grid_xx.max() - grid_xx.min()) * 0.1: # 10%的范围内
|
||||
grid_content[i, j] = values[nearest_idx]
|
||||
|
||||
valid_count = np.sum(~np.isnan(grid_content))
|
||||
print(f"简化插值完成，有效点数: {valid_count}")
|
||||
|
||||
if valid_count > 0:
|
||||
return grid_content
|
||||
else:
|
||||
raise ValueError("简化插值也没有产生有效结果")
|
||||
|
||||
except Exception as e3:
|
||||
print(f"简化插值失败: {e3}")
|
||||
|
||||
print("尝试立方插值作为最后手段...")
|
||||
try:
|
||||
# 最后尝试立方插值
|
||||
grid_content = griddata(
|
||||
points, values, (grid_xx, grid_yy),
|
||||
method='cubic', fill_value=np.nan
|
||||
)
|
||||
|
||||
# 如果立方插值有NaN，用最近邻填充
|
||||
if np.any(np.isnan(grid_content)):
|
||||
print("用最近邻插值填充立方插值的NaN值...")
|
||||
grid_nearest = griddata(
|
||||
points, values, (grid_xx, grid_yy),
|
||||
method='nearest'
|
||||
)
|
||||
nan_mask = np.isnan(grid_content)
|
||||
grid_content[nan_mask] = grid_nearest[nan_mask]
|
||||
|
||||
valid_count = np.sum(~np.isnan(grid_content))
|
||||
print(f"立方插值成功，有效点数: {valid_count}")
|
||||
return grid_content
|
||||
|
||||
except Exception as e4:
|
||||
print(f"立方插值也失败: {e4}")
|
||||
print(f"所有插值方法都失败")
|
||||
raise ValueError("无法完成空间插值，请检查数据点的分布和数值")
|
||||
tree = cKDTree(points)
|
||||
k = min(n_neighbors, len(points))
|
||||
distances, indices = tree.query(grid_flat, k=k)
|
||||
# 防止距离为 0 的除零
|
||||
distances = np.maximum(distances, 1e-12)
|
||||
weights = 1.0 / (distances ** power)
|
||||
# 归一化权重
|
||||
weights /= weights.sum(axis=1, keepdims=True)
|
||||
# 加权求和
|
||||
neighbor_vals = values_flat[indices] if k == 1 else values_flat[indices]
|
||||
if k == 1:
|
||||
result = neighbor_vals
|
||||
else:
|
||||
result = np.sum(weights * neighbor_vals, axis=1)
|
||||
return result.reshape(grid_shape)
|
||||
|
||||
def read_csv_data(self, csv_file, uncertainty_col=None):
|
||||
"""
|
||||
@ -2514,28 +2517,30 @@ class ContentMapper:
|
||||
safe_h = min(float(figsize[1]), _max_inch)
|
||||
fig, ax = plt.subplots(figsize=(safe_w, safe_h))
|
||||
|
||||
# 计算有效值统计（2σ 标准差拉伸，排除长尾异常值干扰）
|
||||
valid = array[~np.isnan(array)]
|
||||
# 1. 明确排除 NaN 以及 nodata_value（与函数参数保持一致）
|
||||
nodata_val = nodata_value
|
||||
valid = array[(~np.isnan(array)) & (array != nodata_val)]
|
||||
|
||||
if valid.size == 0:
|
||||
raise ValueError("GeoTIFF 中没有有效数据（全部为 NoData）")
|
||||
|
||||
mean_val = float(np.nanmean(array))
|
||||
std_val = float(np.nanstd(array))
|
||||
vmin = max(float(np.nanmin(array)), mean_val - 2 * std_val)
|
||||
vmax = min(float(np.nanmax(array)), mean_val + 2 * std_val)
|
||||
# 2. 改用更鲁棒的 2%-98% 百分位拉伸（抗偏态分布）
|
||||
vmin = float(np.percentile(valid, 2))
|
||||
vmax = float(np.percentile(valid, 98))
|
||||
|
||||
if (vmax - vmin) < 1e-9:
|
||||
center = mean_val
|
||||
if (vmax - vmin) < 1e-9: # 防退化：区间过窄 → 取中心 ±1%
|
||||
center = vmin
|
||||
exp = max(abs(center) * 0.01, 1e-9)
|
||||
vmin = center - exp
|
||||
vmax = center + exp
|
||||
|
||||
print(f"[visualize_raster] 2σ 拉伸: vmin={vmin:.4f}, vmax={vmax:.4f}，"
|
||||
f"mean={mean_val:.4f}, std={std_val:.4f}，有效像元: {valid.size}/{array.size}")
|
||||
print(f"[visualize_raster] P2-P98 拉伸: vmin={vmin:.4f}, vmax={vmax:.4f}，"
|
||||
f"有效像元: {valid.size}/{array.size}")
|
||||
|
||||
# ── 栅格绘图 ─────────────────────────────────────────────────
|
||||
# 使用 masked array：NaN 区域自动不显示
|
||||
masked_data = np.ma.masked_invalid(array)
|
||||
# 不仅要屏蔽 NaN，如果 array 中还有 nodata_value，也必须 mask 掉，否则出图时背景会被渲染
|
||||
masked_data = np.ma.masked_where((np.isnan(array)) | (array == nodata_val), array)
|
||||
|
||||
# 【核心修复2】废弃错误的坐标映射逻辑。
|
||||
# 直接使用原生的 imshow，明确告知 matplotlib 第0行在最上方(origin='upper')
|
||||
|
||||
Reference in New Issue
Block a user