这里我们使用的商品数据集，其中包含零售柜中常见的113种零售商品，比如奶茶、方便面、可乐、果汁等各种商品，其类别如下：

Chinese_name = {"3+2-2": "3+2-2", "3jia2": "3+2", "aerbeisi": "阿尔卑斯", "anmuxi": "安慕希", "aoliao": "奥利奥",                "asamu": "阿萨姆", "baicha": "白茶", "baishikele": "百事可乐", "baishikele-2": "百事可乐-2",                "baokuangli": "宝矿力", "binghongcha": "冰红茶", "bingqilinniunai": "冰淇淋牛奶",                "bingtangxueli": "冰糖雪梨", "buding": "布丁", "chacui": "茶萃", "chapai": "茶π", "chapai2": "茶π-2",                "damaicha": "大麦茶", "daofandian1": "到饭点-1", "daofandian2": "到饭点-2", "daofandian3": "到饭点-3",                "daofandian4": "到饭点-4", "dongpeng": "东鹏特饮", "dongpeng-b": "东鹏特饮-b", "fenda": "芬达",                "gudasao": "顾大嫂", "guolicheng": "果粒橙", "guolicheng2": "果粒橙-2", "haitai": "海苔",                "haochidian": "好吃点", "haoliyou": "好丽友", "heweidao": "合味道", "heweidao2": "合味道-2",                "heweidao3": "合味道-3", "hongniu": "红牛", "hongniu2": "红牛-2", "hongshaoniurou": "红烧牛肉",                "jianjiao": "尖叫", "jianlibao": "健力宝", "jindian": "金典", "kafei": "咖啡", "kaomo_gali": "烤馍-咖喱",                "kaomo_jiaoyan": "烤馍-椒盐", "kaomo_shaokao": "烤馍-烧烤", "kaomo_xiangcon": "烤馍-香葱", "kebike": "可比克",                "kele": "可乐", "kele-b": "可乐-b", "kele-b-2": "可乐-b-2", "laotansuancai": "老坛酸菜",                "liaomian": "撩面", "libaojian": "力保健", "lingdukele": "零度可乐", "lingdukele-b": "零度可乐-b",                "liziyuan": "李子园", "lujiaoxiang": "鹿角巷",                "lujikafei": "露吉咖啡", "luxiangniurou": "卤香牛肉", "maidong": "脉动", "mangguoxiaolao": "芒果小酪",                "meiniye": "梅尼耶", "mengniu": "蒙牛",                "mengniuzaocan": "蒙牛早餐奶",                "moliqingcha": "茉莉清茶", "nfc": "NFC果汁", "niudufen": "牛肚粉", "niunai": "牛奶",                "nongfushanquan": "农夫山泉", "qingdaowangzi-1": "青岛王子-1",                "qingdaowangzi-2": "青岛王子-2",                "qinningshui": "沁柠水", "quchenshixiangcao": "屈臣氏香草味苏打水", "rancha-1": "燃茶", "rancha-2": "燃茶",                "rousongbing": "肉松饼",                "rusuanjunqishui": "乳酸菌汽水",                "suanlafen": "酸辣粉",                "suanlaniurou": "酸辣牛肉", "taipingshuda": "太平梳打", "tangdaren": "汤达人", "tangdaren2": "汤达人-2",                "tangdaren3": "汤达人-3", "ufo": "UFO",                "ufo2": "UFO-2",                "wanglaoji": "王老吉", "wanglaoji-c": "王老吉-c",                "wangzainiunai": "旺仔牛奶", "weic": "维C", "weitanai": "维他奶", "weitanai2": "维他奶-2",                "weitanaiditang": "维他奶低糖",                "weitaningmeng": "维他柠檬", "weitaningmeng-bottle": "维他柠檬-瓶装",                "weiweidounai": "维他豆奶", "wuhounaicha": "午后奶茶", "wulongcha": "乌龙茶", "xianglaniurou": "香辣牛肉",                "xianguolao": "鲜果捞",                "xianxiayuban": "鲜虾鱼板面", "xuebi": "雪碧", "xuebi-b": "雪碧-b",                "xuebi2": "雪碧-2", "yezhi": "椰汁", "yibao": "怡宝", "yida": "益达", "yingyangkuaixian": "营养快线",                "yitengyuan": "伊藤园",                "youlemei": "优乐美", "yousuanru": "优酸乳", "youyanggudong": "有养谷咚",                "yuanqishui": "元气水", "zaocanmofang": "早餐魔方", "zihaiguo": "自嗨锅"}

该数据集中，训练集有3796张图片，验证集1084张图片，测试集542张图片，共计5422张图片，选取部分数据部分样本数据集如图所示。

每张图像均提供了图像类标记信息，图像中零售柜商品的bounding box，零售柜商品的关键part信息，以及零售柜商品的属性信息，数据集并解压后得到如下的图片

对于原始的数据集标签文件格式不一致的问题，由于yolov5只认txt而不认json，因此还要有一个转换的过程：批量json转txt。可采用如下代码：

import jsonimport os name2id = {"stitches": 0}  # 标签名称  def convert(img_size, box):    dw = 1. / (img_size[0])    dh = 1. / (img_size[1])    x = (box[0] + box[2]) / 2.0 - 1    y = (box[1] + box[3]) / 2.0 - 1    w = box[2] - box[0]    h = box[3] - box[1]    x = x * dw    w = w * dw    y = y * dh    h = h * dh    return (x, y, w, h)  def decode_json(json_floder_path, json_name):    txt_name = "/mnt/data/yolov5-master/nxm_data/labels_txt/" + json_name[0:-5] + ".txt"    # 存放txt的绝对路径    txt_file = open(txt_name, "w")     json_path = os.path.join(json_floder_path, json_name)    data = json.load(open(json_path, "r", encoding="gb2312", errors="ignore"))     img_w = data["imageWidth"]    img_h = data["imageHeight"]     for i in data["shapes"]:         label_name = i["label"]        if (i["shape_type"] == "rectangle"):            x1 = int(i["points"][0][0])            y1 = int(i["points"][0][1])            x2 = int(i["points"][1][0])            y2 = int(i["points"][1][1])             bb = (x1, y1, x2, y2)            bbox = convert((img_w, img_h), bb)            txt_file.write(str(name2id[label_name]) + " " + " ".join([str(a) for a in bbox]) + "\n")  if __name__ == "__main__":     json_floder_path = "/mnt/data/yolov5-master/nxm_data/labels/"    # 存放json的文件夹的绝对路径    json_names = os.listdir(json_floder_path)    for json_name in json_names:        decode_json(json_floder_path, json_name)

YoloV5原作者给出了4种模型的配置，分别为YoloV5s，YoloV5m，YoloV5l，YoloV5x，根据根据自己的实时性需求选择对应的网络。

接下来只需要在train.py中修改对应的配置文件就好了，对于本数据集配置的yaml文件内容如下：。

train: ./Goods/images/trainval: ./Goods/images/valtest: ./Goods/images/testnc: 113names: ["3+2-2", "3jia2", "aerbeisi", "anmuxi", "aoliao", "asamu", "baicha","baishikele", "baishikele-2","baokuangli",        "binghongcha", "bingqilinniunai", "bingtangxueli", "buding", "chacui", "chapai", "chapai2", "damaicha",        "daofandian1", "daofandian2", "daofandian3", "daofandian4", "dongpeng", "dongpeng-b", "fenda", "gudasao",        "guolicheng", "guolicheng2", "haitai", "haochidian", "haoliyou", "heweidao", "heweidao2", "heweidao3",        "hongniu", "hongniu2", "hongshaoniurou", "jianjiao", "jianlibao", "jindian", "kafei", "kaomo_gali",        "kaomo_jiaoyan", "kaomo_shaokao", "kaomo_xiangcon", "kebike", "kele", "kele-b", "kele-b-2",        "laotansuancai", "liaomian", "libaojian", "lingdukele","lingdukele-b","liziyuan","lujiaoxiang",        "lujikafei","luxiangniurou","maidong","mangguoxiaolao","meiniye","mengniu","mengniuzaocan",        "moliqingcha","nfc","niudufen","niunai","nongfushanquan","qingdaowangzi-1","qingdaowangzi-2",        "qinningshui","quchenshixiangcao","rancha-1","rancha-2","rousongbing","rusuanjunqishui","suanlafen",        "suanlaniurou","taipingshuda","tangdaren","tangdaren2","tangdaren3","ufo","ufo2","wanglaoji","wanglaoji-c",        "wangzainiunai","weic", "weitanai","weitanai2","weitanaiditang","weitaningmeng","weitaningmeng-bottle",        "weiweidounai","wuhounaicha","wulongcha","xianglaniurou","xianguolao","xianxiayuban","xuebi","xuebi-b",        "xuebi2","yezhi","yibao","yida","yingyangkuaixian","yitengyuan","youlemei","yousuanru","youyanggudong",        "yuanqishui","zaocanmofang","zihaiguo"]

（1）train 指定训练集图像路径

（2）val 指定验证集图像路径

（3）test 指定测试集图像路径

（4）nc 指定目标类别数量

（5）目标对应类别名称

以上就是YOLOv5的整体介绍，接下来进行训练。本项目使用Yolov5训练了一个商品目标检测模型，在笔记本的3070显卡下训练了300 epoch，训练过程的截图如下：

在我们的训练过程中，mAP50作为一种常用的目标检测评估指标很快达到了较高水平，而mAP50:95也在训练的过程中不断提升，说明我们模型从训练-验证的角度表现良好。读入一个测试文件夹进行预测，通过训练得到的选取验证集上效果最好的权重best.pt进行实验，得到PR曲线如下图所示。

在深度学习中，我们通常通过损失函数下降的曲线来观察模型训练的情况。而YOLOv5训练时主要包含三个方面的损失：矩形框损失(box_loss)、置信度损失（obj_loss）和分类损失(cls_loss)，在训练结束后，我们也可以在logs目录下找到生成对若干训练过程统计图。下图为博主训练商品识别的模型训练曲线图。

以PR-curve为例，可以看到我们的模型在验证集上的均值平均准确率为0.971。

3.智能零售柜商品检测

def predict(img):    img = torch.from_numpy(img).to(device)    img = img.half() if half else img.float()    img /= 255.0    if img.ndimension() == 3:        img = img.unsqueeze(0)    t1 = time_synchronized()    pred = model(img, augment=False)[0]    pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes,                               agnostic=opt.agnostic_nms)    t2 = time_synchronized()    InferNms = round((t2 - t1), 2)    return pred, InferNms

mainWindow = :/images/icons/back-image.png

mainWindow = ./icons/background2.png