Python项目实战篇——常用验证码标注&识别(CNN神经网络模型训练/测试/部署)

**数学基础**[微积分](https://www.bilibili.com/video/BV1Eb411u7Fw)
[线性代数](https://www.bilibili.com/video/BV1aW411Q7x1)
[概率论](https://www.bilibili.com/video/BV1ot411y7mU)
[计算机数学基础](https://www.bilibili.com/video/BV1AB4y1K7kM)

**OpenCV**[OpenCV文章专栏](https://blog.csdn.net/yukinoai/category_9283880.html)
[OpenCV-Python视频](https://www.bilibili.com/video/BV1tb4y1C7j7)

**神经网络**[理解卷积意义](https://www.bilibili.com/video/BV1VV411478E)
[前馈神经网络](https://www.bilibili.com/video/BV1Tt411s7fK)
[神经网络学习理解](https://space.bilibili.com/504715181?spm_id_from=333.788.b_765f7570696e666f.1)

**Python框架使用**[Numpy中文教程](https://www.runoob.com/numpy/numpy-tutorial.html)
[PyTorch中文教程](https://pytorch.panchuang.net/SecondSection/neural_networks/)
[PyTorch视频](https://www.bilibili.com/video/BV1t64y1t7V8)

if __name__ == '__main__':    split_image_dir(SRC_IMG_DIR)        split_test_image()

labels = []#2-9for i in range(8):    labels.append(50 + i)#A-Z    for i in range(26):    labels.append(65 + i)
class VerCodeDataset(Dataset):    def __init__(self, image_dir="./letter_template/"):        l = os.listdir(image_dir)        self.data = []        self.label = []        for d in l:            fs = os.listdir("{}{}".format(image_dir, d))            for f in fs:                fup = "{}{}/{}".format(image_dir, d, f)                #图片numpy转tensor                t = torch.from_numpy(io.imread(fup)).float() / 255                #将二维值标准化                norl = transforms.Normalize(t.mean(), t.std())                self.data.append(norl(t.reshape(1, 40, 40)))                #添加字符对应标签序号                self.label.append(labels.index(ord(d)))

[数据集制作参考文章](https://zhuanlan.zhihu.com/p/358671390)

[PyTorch 神经网络 - PyTorch官方教程中文版](https://link.zhihu.com/?target=http%3A//pytorch.panchuang.net/SecondSection/neural_networks/)

class Net(nn.Module):    def __init__(self, dropout=0.1):        super(Net, self).__init__()        self.dropout = nn.Dropout(dropout)         #第一层，卷积核个数从6改成10        self.conv1 = nn.Conv2d(1, 10, 5)        #第二层，卷积核个数从10改成25        self.conv2 = nn.Conv2d(10, 25, 5)                #全连接层1，40*40的字符图经过2层卷积+2层池化变成7*7        self.fc1 = nn.Linear(1 * 25 * 7 * 7, 120)        #全连接层2        self.fc2 = nn.Linear(120, 84)        #最后全连接3层为输出层，本案例验证码分类一共34类，[2-9,A-Z],改为34。        self.fc3 = nn.Linear(84, 34)
    def forward(self, x):        # 池化出来大小直接除2        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))        #防止过拟合        x = self.dropout(x)        x = F.max_pool2d(F.relu(self.conv2(x)), (2, 2))        x = self.dropout(x)        x = x.view(-1, self.num_flat_features(x))        #神经元relu激活函数        x = F.relu(self.fc1(x))        x = F.relu(self.fc2(x))        x = self.fc3(x)        return x
    def num_flat_features(self, x):        size = x.size()[1:]  # all dimensions except the batch dimension        num_features = 1        for s in size:            num_features *= s        return num_features

self.fc1 = nn.Linear(1 * 25 * 7 * 7, 120)

数据量：2286张 40*40 单通道字符图片batch_size: 50epoch: 200
设备          时间GTX 1070TI      25sAMD R7 4750U PRO  4min

[cuda安装文章链接](https://www.cnblogs.com/yang520ming/p/10677110.html)

1.更新nvida显卡驱动程序，然后看cuda版本2.找pytorch对应cuda的版本安装

def train_gpu():    use_cuda = torch.cuda.is_available()    if(use_cuda):        print("use gpu cuda")    else:        print("use cpu")
    device = torch.device("cuda:0" if use_cuda else "cpu")    net = Net()    net.to(device)    #随机梯度下降    opt = optim.SGD(net.parameters(), lr=0.01)    #迭代数据200次    epoch = 200    #单批次数据为50个    batch_size = 50        trainloader = data.trainloader(batch_size)    st = datetime.datetime.now()    loss = 0
    for e in range(epoch):        for step, d in enumerate(trainloader):            data_cuda =  d["data"].to(device)            label_cuda = d["label"].to(device)            #每次反向传播后，梯度清零            opt.zero_grad()            #前向传播            out = net(data_cuda)            #分类问题选用交叉熵损失函数            lf = nn.CrossEntropyLoss()            #计算损失            loss = lf(out, label_cuda)            #反向传播修改神经元参数            loss.backward()            opt.step()
      #每迭代50次或第一个批次步骤数据输出损失值            if (e % 50 == 0 or step == 1):                print("e : {} , step : {}, loss : {}".format(e, step, loss))
    print("loss : {}".format(loss))    #输出训练时间    print("cost time:",datetime.datetime.now() - st)    #保存模型    saveModel(net, opt)

[神经网络学习理解](https://space.bilibili.com/504715181?spm_id_from=333.788.b_765f7570696e666f.1)

@app.route('/recognize/path', methods=['POST'])def recognize_path():    filePathList = request.json['filePathList']    code = CODE_SUCCESS    msg = MSG_SUCCESS    data = []    for filePath in filePathList:        if not os.path.exists(filePath):            # code = CODE_FAIL            # msg = "文件不存在"            print("文件不存在:", filePath)            data.append("")            continue        else:            labels = usage_model.usage(filePath)            data.append(''.join(labels))    result = {'code': code, "msg": msg, "data": data}    return jsonify(result)