版本变更V3.31添加识图功能

inventory-backend/app/api/v1/common/image_search.py

@@ -80,73 +80,81 @@ def image_search():
             SELECT id, name, spec_model, image_url,
                    (1 - (vec <=> :query_vector)) AS similarity
             FROM (
-                SELECT id,
-                       COALESCE(name, '') AS name,
-                       COALESCE(spec, '') AS spec_model,
-                       COALESCE(product_image, '') AS image_url,
-                       img_embedding AS vec
+                -- 1. 基础物料表
+                SELECT id, name, spec_model, product_image AS image_url, img_embedding AS vec
                 FROM material_base
                 WHERE img_embedding IS NOT NULL
 
                 UNION ALL
 
-                SELECT id,
-                       '采购入库' AS name,
-                       '到货照片' AS spec_model,
-                       COALESCE(arrival_photo, '') AS image_url,
-                       arrival_image_embedding AS vec
-                FROM stock_buy
-                WHERE arrival_image_embedding IS NOT NULL
+                -- 2. 采购入库表 (通过 base_id 关联拿真实物料)
+                SELECT mb.id, mb.name, mb.spec_model, sb.arrival_photo AS image_url, sb.arrival_image_embedding AS vec
+                FROM stock_buy sb
+                JOIN material_base mb ON sb.base_id = mb.id
+                WHERE sb.arrival_image_embedding IS NOT NULL
 
                 UNION ALL
 
-                SELECT id,
-                       '采购入库' AS name,
-                       '质检报告' AS spec_model,
-                       COALESCE(qc_report, '') AS image_url,
-                       qc_report_image_embedding AS vec
-                FROM stock_buy
-                WHERE qc_report_image_embedding IS NOT NULL
+                -- 3. 半成品入库表
+                SELECT mb.id, mb.name, mb.spec_model, ss.arrival_photo AS image_url, ss.arrival_image_embedding AS vec
+                FROM stock_semi ss
+                JOIN material_base mb ON ss.base_id = mb.id
+                WHERE ss.arrival_image_embedding IS NOT NULL
+
+                UNION ALL
+
+                -- 4. 成品入库表
+                SELECT mb.id, mb.name, mb.spec_model, sp.product_photo AS image_url, sp.arrival_image_embedding AS vec
+                FROM stock_product sp
+                JOIN material_base mb ON sp.base_id = mb.id
+                WHERE sp.arrival_image_embedding IS NOT NULL
             ) AS combined
-            ORDER BY vec <=> :query_vector
-            LIMIT 10
+            ORDER BY vec <=> :query_vector LIMIT 10
         """)
 
-        result = db.session.execute(sql, {"query_vector": query_vector_str})
-        rows = result.fetchall()
+        # 执行查询
+        records = db.session.execute(sql, {"query_vector": query_vector_str}).fetchall()
 
         results = []
-        for row in rows:
-            item_id = row[0]
-            item_name = row[1] or ""
-            spec_model = row[2] or ""
-            raw_image = row[3]
+        seen_product_ids = set()  # 【新增】用来记录已经添加过的物料 ID
 
-            # 解析图片 URL 列表，取第一张
-            image_url = ""
-            if raw_image:
-                try:
-                    image_list = json.loads(raw_image)
-                    if image_list and len(image_list) > 0:
-                        image_url = image_list[0]
-                except Exception:
-                    # 纯字符串直接使用
-                    image_url = str(raw_image)
+        for row in records:
+            # 【新增】如果这个物料已经在这个列表里了，直接跳过它
+            if row.id in seen_product_ids:
+                continue
 
+            # 记录这个物料 ID，保证下次不会再重复添加
+            seen_product_ids.add(row.id)
+
+            # 1. 提取原始 URL
+            raw_url = row.image_url
+            clean_url = ""
+
+            if raw_url:
+                if raw_url.startswith('[') and raw_url.endswith(']'):
+                    import json
+                    try:
+                        url_list = json.loads(raw_url)
+                        clean_url = url_list[0] if url_list else ""
+                    except:
+                        clean_url = raw_url
+                else:
+                    clean_url = raw_url
+
+            # 2. 组装返回结果
             results.append({
-                "id": item_id,
-                "name": item_name,
-                "spec_model": spec_model,
-                "image_url": image_url,
-                "similarity": round(float(row[4]), 4)
+                "product_id": row.id,
+                "product_name": row.name,
+                "spec_model": row.spec_model,
+                "image_url": clean_url,
+                "similarity": round(float(row.similarity), 4)
             })
 
-        print(f"✅ [ImageSearch] 跨表检索完成，命中 {len(results)} 条结果")
-        return jsonify({
-            "code": 200,
-            "msg": "检索成功",
-            "data": results
-        })
+            # 【新增】只要凑够了 10 个完全不同的物料，就立刻结束循环
+            if len(results) >= 10:
+                break
+
+        return jsonify({"code": 200, "data": results})
 
     except Exception as e:
         print(f"❌ [ImageSearch] 数据库检索失败: {e}")

inventory-backend/app/services/inbound/base_service.py

@@ -1048,14 +1048,15 @@ class MaterialBaseService:
     @staticmethod
     def get_latest_specs():
         """
-        获取所有规格型号的最大连号，按连续区间分组返回
+        获取所有规格型号的分组统计，按规则聚合后返回
         - 前缀统一大写处理
-        - 只有数字完全连续（N, N+1, N+2...）才认定为同一组
-        - 数字不连续时断开，形成新组
-        - 按每组数量降序排列
-        - 返回每个连续区间的最大值
+        - 匹配模式：(前缀)(单数字二级分类位)(纯数字部分)，如 OPT12046 -> OPT, 1, 2046
+        - OPT 系列：使用 前缀+二级分类位 作为分组 Key，如 OPT1, OPT2
+        - 其他前缀：直接使用前缀作为分组 Key
+        - 返回每个分组的数量、最大号、完整规格名
         """
         import re
+        from collections import defaultdict
 
         # 1. 查询所有不为空的规格型号
         specs = MaterialBase.query.filter(
@@ -1063,8 +1064,8 @@ class MaterialBaseService:
             MaterialBase.spec_model != ''
         ).all()
 
-        # 2. 解析并收集所有有效的 (prefix, num, original_spec)
-        parsed = []
+        # 2. 按分组收集所有数字
+        groups = defaultdict(list)
 
         for material in specs:
             spec = material.spec_model
@@ -1072,72 +1073,31 @@ class MaterialBaseService:
                 continue
 
             base_spec = spec.split('/')[0]
-
-            match = re.match(r'^([A-Za-z]+)(\d+)$', base_spec)
+            match = re.match(r'^([A-Za-z]+)(\d)(\d+)$', base_spec)
             if not match:
                 continue
 
-            prefix, num_str = match.groups()
+            prefix, sub_cat, num_str = match.groups()
             prefix = prefix.upper()
             num = int(num_str)
 
-            parsed.append((prefix, num, spec))
+            # OPT 系列使用 前缀+单数字二级分类 作为 Key
+            key = f"{prefix}{sub_cat}" if prefix == 'OPT' else prefix
+            groups[key].append((num, spec))
 
-        # 3. 先按 prefix 升序，再按 num 升序排序
-        parsed.sort(key=lambda x: (x[0], x[1]))
-
-        # 4. 遍历切分连续区间
-        # 核心逻辑：当 current_num != prev_num + 1 时，断开形成新组
-        intervals = []
-        current_prefix = None
-        current_start = None
-        current_end = None
-        current_last_spec = None
-
-        for prefix, num, spec in parsed:
-            if current_prefix is None:
-                current_prefix = prefix
-                current_start = num
-                current_end = num
-                current_last_spec = spec
-            elif prefix == current_prefix and num == current_end + 1:
-                current_end = num
-                current_last_spec = spec
-            else:
-                intervals.append({
-                    'prefix': current_prefix,
-                    'start': current_start,
-                    'end': current_end,
-                    'count': current_end - current_start + 1,
-                    'latest': current_last_spec
-                })
-                current_prefix = prefix
-                current_start = num
-                current_end = num
-                current_last_spec = spec
-
-        if current_prefix is not None:
-            intervals.append({
-                'prefix': current_prefix,
-                'start': current_start,
-                'end': current_end,
-                'count': current_end - current_start + 1,
-                'latest': current_last_spec
-            })
-
-        # 5. 按每组数量降序排列，再按前缀升序
-        intervals.sort(key=lambda x: (-x['count'], x['prefix']))
-
-        # 6. 构建返回结果
+        # 3. 生成展示用的统计数据
         result = []
-        for item in intervals:
-            prefix = item['prefix']
-            start = item['start']
-            end = item['end']
+        for key, items in groups.items():
+            sorted_items = sorted(items, key=lambda x: x[0])
+            max_num, max_spec = sorted_items[-1]
             result.append({
-                "group": f"{prefix}({start}-{end})",
-                "count": item['count'],
-                "latest": item['latest']
+                'group': key,
+                'count': len(sorted_items),
+                'latest': max_spec,
+                'max_num': max_num
             })
 
+        # 4. 按数量降序，再按分组名升序排列
+        result.sort(key=lambda x: (-x['count'], x['group']))
+
         return result

inventory-backend/requirements.txt

@@ -25,4 +25,6 @@ openpyxl>=3.1.2
 # [新增] 定时任务调度器 (库存预警每日邮件)
 APScheduler==3.10.4
 # [新增] 时区处理 (APScheduler 需要)
-pytz
+pytz
+# [新增] 进度条库 (脚本和任务所需)
+tqdm>=4.66.0

inventory-backend/scripts/init_all_vectors.py

@@ -26,15 +26,26 @@ from app.extensions import db
 from app.utils.ai_vision import get_image_embedding, load_clip_model
 
 # ============================================================================
-# 业务配置：表 → 图片字段 → 向量字段 映射
+# 业务配置：表 → 图片字段 → 向量字段 映射 (已全面修复)
 # ============================================================================
 TARGET_TABLES = [
-    # 基础物料
+    # 1. 基础物料
     {"table": "material_base", "img_col": "product_image", "vec_col": "img_embedding"},
 
-    # 采购入库
+    # 2. 采购入库
     {"table": "stock_buy", "img_col": "arrival_photo", "vec_col": "arrival_image_embedding"},
-    {"table": "stock_buy", "img_col": "qc_report", "vec_col": "qc_report_image_embedding"},
+    {"table": "stock_buy", "img_col": "inspection_report", "vec_col": "qc_report_image_embedding"}, # 已修复: qc_report -> inspection_report
+
+    # 3. 半成品入库 (新增)
+    {"table": "stock_semi", "img_col": "arrival_photo", "vec_col": "arrival_image_embedding"},
+    {"table": "stock_semi", "img_col": "quality_report_link", "vec_col": "qc_report_image_embedding"},
+
+    # 4. 成品入库 (新增)
+    {"table": "stock_product", "img_col": "product_photo", "vec_col": "arrival_image_embedding"},
+    {"table": "stock_product", "img_col": "quality_report_link", "vec_col": "qc_report_image_embedding"}
+
+    # 注意：成品入库表还有一个 inspection_report_link，但由于数据库中成品表目前只加了两个向量字段，
+    # 暂不将该字段加入遍历，以免覆盖 quality_report_link 的特征。
 ]
 
 # 物理图片根目录（相对于 app 目录的相对路径 ../uploads/）