恢复使用图像识别仓库物品

auto_get_res_new.py

@@ -441,7 +441,7 @@ class Arknights数据处理器:
     def 批量训练并保存扫仓库模型(self):
         self.训练仓库的knn模型("NORMAL", "./mower/models/NORMAL.pkl")
         self.训练仓库的knn模型("CONSUME", "./mower/models/CONSUME.pkl")
-        # self.训练仓库的knn模型("MATERIAL", "./mower/models/MATERIAL.pkl")
+        self.训练仓库的knn模型("MATERIAL", "./mower/models/MATERIAL.pkl")
 
     def 训练在房间内的干员名的模型(self):
         font = ImageFont.truetype("mower/fonts/SourceHanSansCN-Medium.otf", 37)

mower/solvers/base_schedule.py

@@ -10,7 +10,8 @@ from typing import Literal
 from mower.data import agent_list, base_room_list
 from mower.solvers.credit import CreditSolver
 from mower.solvers.credit_fight import CreditFight
-from mower.solvers.cultivate_depot import cultivate as cultivateDepotSolver
+
+# from mower.solvers.cultivate_depot import cultivate as cultivateDepotSolver
 from mower.solvers.depotREC import depotREC as DepotSolver
 from mower.solvers.infra.base_mixin import BaseMixin
 from mower.solvers.infra.clue import ClueSolver
@@ -2866,7 +2867,7 @@ class BaseSchedulerSolver(SceneGraphSolver, BaseMixin):
 
     def 仓库扫描(self):
         try:
-            cultivateDepotSolver().start()
+            # cultivateDepotSolver().start()
             DepotSolver().run()
         except Exception as e:
             logger.exception(f"先不运行 出bug了 : {e}")

mower/solvers/depotREC.py

@@ -9,6 +9,7 @@ import cv2
 import numpy as np
 import pandas as pd
 from skimage.feature import hog
+from sklearn.cluster import KMeans
 
 from mower import __rootdir__
 from mower.utils import config
@@ -36,6 +37,47 @@ def 导入_数字模板():
     return 数字模板列表
 
 
+def 找几何中心(coordinates, n_clusters=3):
+    coordinates_array = np.array(coordinates).reshape(-1, 1)
+    kmeans = KMeans(n_clusters=n_clusters)
+    kmeans.fit(coordinates_array)
+    centers = kmeans.cluster_centers_.flatten().astype(int)
+    logger.debug(centers)
+
+    return sorted(centers)
+
+
+def 找圆(拼接结果, 参数1=50, 参数2=30, 圆心间隔=230, 最小半径=90, 最大半径=100):
+    灰图 = cv2.cvtColor(拼接结果, cv2.COLOR_RGB2GRAY)
+    圆 = cv2.HoughCircles(
+        灰图,
+        cv2.HOUGH_GRADIENT,
+        dp=1,
+        minDist=圆心间隔,
+        param1=参数1,
+        param2=参数2,
+        minRadius=最小半径,
+        maxRadius=最大半径,
+    )
+
+    return 圆
+
+
+def 拼图(图片列表):
+    try:
+        stitcher = cv2.Stitcher.create(mode=cv2.Stitcher_SCANS)
+        status, result = stitcher.stitch(图片列表)
+        if status == cv2.Stitcher_OK:
+            logger.info("仓库扫描: 拼接完成。")
+            return result
+        else:
+            logger.warning(f"仓库扫描: 拼接失败，状态码: {status}")
+            raise RuntimeError(f"拼接失败，状态码: {status}")
+    except RuntimeError as e:
+        logger.error(f"仓库扫描: 拼接过程中出现错误: {e}")
+        raise
+
+
 def 提取特征点(模板):
     模板 = 模板[40:173, 40:173]
     hog_features = hog(
@@ -50,15 +92,35 @@ def 提取特征点(模板):
     return hog_features
 
 
+def 算坐标(圆):
+    circles_x = sorted(圆[:, 0])
+    groups = []
+    start_index = 0
+
+    for i in range(1, len(circles_x)):
+        if circles_x[i] - circles_x[i - 1] >= 20:
+            groups.append(circles_x[start_index:i])
+            start_index = i
+
+    groups.append(circles_x[start_index:])
+    circles_x = [int(sum(group) / len(group)) for group in groups]
+    circles_y = 找几何中心(圆[:, 1], 3)
+    return circles_x, circles_y
+
+
 def 识别空物品(物品灰):
     物品灰 = 物品灰[0:130, 130:260]
     _, 二值图 = cv2.threshold(物品灰, 60, 255, cv2.THRESH_BINARY)
     白像素个数 = cv2.countNonZero(二值图)
     所有像素个数 = 二值图.shape[0] * 二值图.shape[1]
     白像素比值 = int((白像素个数 / 所有像素个数) * 100)
-
+    # saveimg_depot(
+    #     cv2.hconcat([物品灰, 二值图]),
+    #     f"{白像素比值}_{datetime.now().timestamp()}.png",
+    #     "depot_3_empty",
+    # )
     if 白像素比值 > 99:
-        logger.info("删除一次空物品")
+        logger.info("仓库扫描: 删除一次空物品")
 
         return False
     else:
@@ -80,9 +142,18 @@ def 切图(圆心x坐标, 圆心y坐标, 拼接结果, 正方形边长=130):
                 id = str(datetime.now().timestamp())
                 正方形切 = 正方形[26:239, 26:239]
                 图片.append([正方形切, 正方形灰, id])
+                # saveimg_depot(正方形切, id + ".png", "depot_4_name")
+                # saveimg_depot(正方形灰, id + ".png", "depot_4_num")
     return 图片
 
 
+def 经验卡分类(物品):
+    # 图像处理
+    # return 物品名称
+    logger.info("没有写具体内容 ")
+    pass
+
+
 class depotREC(SceneGraphSolver):
     def __init__(self) -> None:
         start_time = datetime.now()
@@ -90,6 +161,8 @@ class depotREC(SceneGraphSolver):
         # sift = cv2.SIFT_create()
         orb = cv2.ORB_create()
         bf = cv2.BFMatcher(cv2.NORM_HAMMING2, crossCheck=True)
+        # flann = cv2.FlannBasedMatcher(
+        #     dict(algorithm=1, trees=2), dict(checks=50))
         self.detector = orb
         self.matcher = bf
 
@@ -97,17 +170,27 @@ class depotREC(SceneGraphSolver):
 
         with lzma.open(f"{__rootdir__}/models/CONSUME.pkl", "rb") as pkl:
             self.knn模型_CONSUME = pickle.load(pkl)
+        with lzma.open(f"{__rootdir__}/models/MATERIAL.pkl", "rb") as pkl:
+            self.knn模型_MATERIAL = pickle.load(pkl)
         with lzma.open(f"{__rootdir__}/models/NORMAL.pkl", "rb") as pkl:
             self.knn模型_NORMAL = pickle.load(pkl)
+        # self.时间模板 = self.导入_时间模板()
         self.物品数字 = 导入_数字模板()
 
         self.结果字典 = {}
+        self.明日方舟工具箱json = {}
 
-        logger.info(f"吟唱用时{datetime.now() - start_time}")
+        logger.info(f"仓库扫描: 吟唱用时{datetime.now() - start_time}")
 
     def 切图主程序(self, 拼接好的图片):
-        横坐标 = [188 + 234 * i for i in range(0, 8)]
-        纵坐标 = [144, 430, 715]
+        圆 = 找圆(拼接好的图片)
+        if 圆 is not None and len(圆[0]) > 2:
+            圆 = np.round(圆[0, :]).astype("int")
+            横坐标, 纵坐标 = 算坐标(圆)
+        else:
+            横坐标 = [188 + 234 * i for i in range(0, 8)]
+            纵坐标 = [144, 430, 715]
+            logger.warning("仓库扫描: 在这个分类下没有找到足够多的圆，使用预设坐标")
         切图列表 = 切图(横坐标, 纵坐标, 拼接好的图片)
         return 切图列表
 
@@ -152,26 +235,29 @@ class depotREC(SceneGraphSolver):
         super().run()
 
     def transition(self) -> bool:
-        logger.info("回到桌面")
+        logger.info("仓库扫描: 回到桌面")
         self.back_to_index()
         if self.scene() == Scene.INDEX:
             self.tap_index_element("warehouse")
-            logger.info("从主界面点击仓库界面")
+            logger.info("仓库扫描: 从主界面点击仓库界面")
 
             time = datetime.now()
             任务组 = [
                 (1200, self.knn模型_CONSUME, "消耗物品"),
                 (1400, self.knn模型_NORMAL, "基础物品"),
+                (1700, self.knn模型_MATERIAL, "养成材料"),
             ]
 
             for 任务 in 任务组:
                 self.tap((任务[0], 70))
                 if not self.find("depot_empty"):
-                    self.分类扫描(任务[1])
-                    logger.info(f"{任务[2]}识别，识别用时{datetime.now() - time}")
+                    self.分类扫描(任务[1], 任务[2])
+                    logger.info(
+                        f"仓库扫描: {任务[2]}识别，识别用时{datetime.now() - time}"
+                    )
                 else:
-                    logger.info("这个分类下没有物品")
-            logger.info(self.结果字典)
+                    logger.info("仓库扫描: 这个分类下没有物品")
+            logger.info(f"仓库扫描: {self.结果字典}")
             result = [
                 int(datetime.now().timestamp()),
                 json.dumps(self.结果字典, ensure_ascii=False),
@@ -194,17 +280,53 @@ class depotREC(SceneGraphSolver):
         similarity = len(matches) / max(len(descriptors1), len(descriptors2))
         return similarity * 100
 
-    def 分类扫描(self, 模型名称):
+    def 分类扫描(self, 模型名称, 分类名称):
         截图列表 = []
         旧的截图 = config.recog.img
         旧的截图 = 旧的截图[140:1000, :]
         截图列表.append(旧的截图)
+        # saveimg(旧的截图, "depot_1_screenshot")
         config.recog.update()
-        拼接好的图片 = 截图列表[0]
+
+        logger.info(f"仓库扫描: 把第{len(截图列表)}页保存进内存中等待识别")
+        if "养成材料" in 分类名称:
+            while True:
+                self.swipe_noinertia((1800, 450), (-1000, 0))  # 滑动
+                config.recog.update()
+                新的截图 = config.recog.img
+                新的截图 = 新的截图[140:1000, :]
+                # saveimg(新的截图, "depot_1_screenshot")
+                相似度 = self.对比截图(截图列表[-1], 新的截图)
+                if 相似度 < 70:
+                    截图列表.append(新的截图)
+                    logger.info(
+                        f"仓库扫描: 把第{len(截图列表)}页保存进内存中等待识别,相似度{相似度}"
+                    )
+                else:
+                    logger.info("仓库扫描: 这大抵是最后一页了")
+                    break
+        logger.info(f"仓库扫描: 截图读取完了,有{len(截图列表)}张截图")
+        logger.info("仓库扫描: 开始计算裁切图像")
+
+        if len(截图列表) > 1:
+            拼接好的图片 = 拼图(截图列表)
+        else:
+            拼接好的图片 = 截图列表[0]
+        # saveimg(拼接好的图片, "depot_2_stitcher")
+
         切图列表 = self.切图主程序(拼接好的图片)
-        logger.info(f"需要识别{len(切图列表)}个物品")
+
+        logger.info(f"仓库扫描: 需要识别{len(切图列表)}个物品")
 
         for [物品, 物品灰, id] in 切图列表:
             [物品名称, 物品数字] = self.匹配物品一次(物品, 物品灰, 模型名称)
+            # saveimg_depot(
+            #     物品,
+            #     f"{id}_{物品名称}_{物品数字}.png",
+            #     "depot_5_result",
+            # )
             logger.debug([物品名称, 物品数字])
+            if "作战记录" in 物品名称:
+                logger.info("对经验卡进行重新识别")
+                经验卡分类(物品) # TODO 
             self.结果字典[物品名称] = self.结果字典.get(物品名称, 0) + 物品数字

mower/utils/depot.py

@@ -17,6 +17,12 @@ def 读取仓库():
         创建json()
     with open(path, "r", encoding="utf-8") as f:
         depotinfo = json.load(f)
+    depotinfo = {
+        "code": 0,
+        "message": "OK",
+        "timestamp": "0",
+        "data": {"items": [{"id": "0", "count": "0"}]},
+    }    
     物品数量 = depotinfo["data"]["items"]
     新物品1 = {
         key_mapping[item["id"]][2]: int(item["count"])
@@ -224,8 +230,8 @@ def 创建json():
     a = {
         "code": 0,
         "message": "OK",
-        "timestamp": "1719065002",
-        "data": {"items": [{"id": "31063", "count": "0"}]},
+        "timestamp": "0",
+        "data": {"items": [{"id": "0", "count": "0"}]},
     }
     with open(path, "w", encoding="utf-8") as f:
         json.dump(a, f)