'commit'

Apps/TeLaiDian/ReadImageKit.py

@@ -140,6 +140,108 @@ class ReadImageKit:
             logger.error(f"VLM 寻找价格入口失败: {e}")
             return {"found": False}
 
+    async def find_price_tab_ocr(self, image_path):
+        """
+        使用 OCR 在详情页顶部标签栏中寻找“价格”标签
+        不再使用 VLM
+        """
+        from Apps.TeLaiDian.Kit import draw_rectangles
+        img = read_image(image_path)
+        if img is None:
+            return {"found": False}
+        
+        h, w = img.shape[:2]
+        reader = get_easyocr_reader(gpu=True)
+        
+        # 限制只在顶部 30% 区域查找标签
+        roi = img[0:int(h*0.3), :]
+        results = reader.read_text(roi)
+        
+        for (quad, text, prob) in results:
+            if '价格' in text and prob > 0.3:
+                # 转换坐标 (加上 ROI 的偏移，这里 y 偏移是 0)
+                res = reader.get_normalized_rect(quad, w, h)
+                center_x = (res[0] + res[2]) // 2
+                center_y = (res[1] + res[3]) // 2
+                
+                # 绘制诊断图
+                actual_p = [int(center_x * w / 1000), int(center_y * h / 1000)]
+                bbox = [int(res[0] * w / 1000), int(res[1] * h / 1000), int(res[2] * w / 1000), int(res[3] * h / 1000)]
+                draw_rectangles(image_path, bboxes=[bbox], click_points=[actual_p])
+                logger.info(f"[OCR] 已定位顶部价格标签并保存诊断图: {image_path.replace('.jpg', '_tab_vl.jpg')}")
+                
+                return {
+                    "found": True,
+                    "reason": f"OCR 定位到‘{text}’",
+                    "point": [center_x, center_y]
+                }
+        
+        logger.warning(f"[OCR] 未能在顶部区域定位到‘价格’标签")
+        return {"found": False}
+
+    async def find_price_tab_ocr(self, image_path):
+        """
+        使用 OCR 在详情页寻找顶部“价格”标签
+        """
+        img = read_image(image_path)
+        if img is None:
+            return {"found": False}
+        h, w = img.shape[:2]
+        
+        from Util.EasyOcrKit import get_easyocr_reader
+        reader = get_easyocr_reader(gpu=True)
+        
+        # 顶部标签栏通常在屏幕上部 10%-30% 区域
+        # 我们先识别所有文字，然后过滤出“价格”
+        results = reader.read_text(img)
+        for (quad, text, prob) in results:
+            if "价格" in text and prob > 0.3:
+                # 获取中心点
+                rect = reader.get_normalized_rect(quad, w, h)
+                center_x = (rect[0] + rect[2]) // 2
+                center_y = (rect[1] + rect[3]) // 2
+                
+                # 校验位置：必须在屏幕上半部分
+                if center_y < 500:
+                    logger.info(f"[OCR] 找到顶部价格标签: '{text}', 置信度: {prob:.4f}, 坐标: [{center_x}, {center_y}]")
+                    return {
+                        "found": True,
+                        "point": [center_x, center_y]
+                    }
+        
+        return {"found": False}
+
+    async def find_price_entrance_ocr(self, image_path):
+        """
+        使用 OCR 在详情页寻找价格入口 (全部时段 >)
+        不再使用 VLM 和 掩码 CV
+        """
+        container_norm = detect_price_info_container_cv(image_path)
+        if container_norm:
+            # 找到矩形框 [x1, y1, x2, y2]
+            x1, y1, x2, y2 = container_norm
+            # 计算中心点
+            center_x = (x1 + x2) // 2
+            center_y = (y1 + y2) // 2
+            
+            # 绘制诊断图 (复用之前的逻辑)
+            img = read_image(image_path)
+            if img is not None:
+                h, w = img.shape[:2]
+                actual_p = [int(center_x * w / 1000), int(center_y * h / 1000)]
+                bbox = [int(x1 * w / 1000), int(y1 * h / 1000), int(x2 * w / 1000), int(y2 * h / 1000)]
+                draw_rectangles(image_path, bboxes=[bbox], click_points=[actual_p])
+                logger.info(f"[OCR] 已定位价格入口并保存诊断图: {image_path.replace('.jpg', '_vl.jpg')}")
+                
+            return {
+                "found": True,
+                "reason": "OCR 精准定位到‘全部时段’文字",
+                "point": [center_x, center_y]
+            }
+        
+        logger.warning(f"[OCR] 未能在页面中定位到‘全部时段’")
+        return {"found": False}
+
     async def find_close_button_vlm(self, image_path):
         """
         使用 VLM 在页面上寻找关闭按钮（用于清理广告弹窗）