├── .gitignore ├── .idea ├── .gitignore ├── ImageCaptchaOCR.iml ├── inspectionProfiles │ └── profiles_settings.xml ├── misc.xml ├── modules.xml └── vcs.xml ├── README.md ├── development ├── __pycache__ │ ├── gen_ImageCaptcha.cpython-38.pyc │ ├── model.cpython-38.pyc │ └── one_hot.cpython-38.pyc ├── gen_ImageCaptcha.py ├── model.py ├── one_hot.py ├── predict.py ├── pth2onnx.py └── server.py ├── requirement.txt └── 英数验证码数据集.zip /.gitignore: -------------------------------------------------------------------------------- 1 | 2 | ./development/pth2onnx.py 3 | ./英数验证码数据集.zip 4 | /dataset/ 5 | /deploy/ 6 | -------------------------------------------------------------------------------- /.idea/.gitignore: -------------------------------------------------------------------------------- 1 | # Default ignored files 2 | /shelf/ 3 | /workspace.xml 4 | # Editor-based HTTP Client requests 5 | /httpRequests/ 6 | # Datasource local storage ignored files 7 | /dataSources/ 8 | /dataSources.local.xml 9 | -------------------------------------------------------------------------------- /.idea/ImageCaptchaOCR.iml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/inspectionProfiles/profiles_settings.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 6 | -------------------------------------------------------------------------------- /.idea/misc.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | -------------------------------------------------------------------------------- /.idea/modules.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/vcs.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | @[TOC](pytorch_CNN英数验证码识别模型训练) 2 | 3 | ### 安卓逆向，JS逆向，图像识别，在线接单，全套源码+部署+算法联系QQ: 27788854，wechat: taisuivip，[telegram: rtais00](https://t.me/rtais00) 4 | ### 微信公众号：R逆向 5 | # 验证码样式 6 | #### 问题案例： 7 |     项目github：[https://github.com/taisuii/ImageCaptchaOCR](https://github.com/taisuii/ImageCaptchaOCR) 8 |     验证码样式通常为通常为英文数字 9 | ![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/93c57be2e23949988de79c8452341434.png#pic_center) 10 | #### 解决方案或思路： 11 |     卷积神经网络，搭建模型，使其输出为5*(26+26+10)=310的张量，解码为英文和数字 12 | # 数据集准备 13 | #### 整理数据集： 14 |     先关注公众号：R逆向，回复：验证码数据集 15 |     用两个文件夹，train存放数据集，test存放测试数据集，内容如下，标签就是图片名字 16 | ![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/347c6358daf2494d95ddb9cbcf90838a.png#pic_center) 17 | #### 自定义数据集： 18 | ```python 19 | class mydatasets(Dataset): 20 | def __init__(self, root_dir): 21 | super(mydatasets, self).__init__() 22 | self.list_image_path = [os.path.join(root_dir, image_name) for image_name in os.listdir(root_dir)] 23 | self.transforms = transforms.Compose([ 24 | transforms.ToTensor() 25 | ]) 26 | 27 | def __getitem__(self, index): 28 | image_path = self.list_image_path[index] 29 | img_ = Image.open(image_path) 30 | image_name = image_path.split("\\")[-1] 31 | img_tesor = self.transforms(img_) 32 | img_lable = image_name.split(".")[0] 33 | img_lable = one_hot.text2vec(img_lable) 34 | img_lable = img_lable.view(1, -1)[0] 35 | return img_tesor, img_lable 36 | 37 | def __len__(self): 38 | return self.list_image_path.__len__() 39 | ``` 40 | # 模型训练 41 | #### 模型代码如下： 42 |     随便写几个层，留意输入和输出，gen_ImageCaptcha.captcha_size * len(gen_ImageCaptcha.captcha_array)就是5*(26+26+10) 43 | ```python 44 | class MyModel(nn.Module): 45 | def __init__(self): 46 | super(MyModel, self).__init__() 47 | self.layer1 = nn.Sequential( 48 | nn.Conv2d(in_channels=3, out_channels=64, kernel_size=3, padding=1), 49 | nn.BatchNorm2d(64), # 添加Batch Normalization 50 | nn.ReLU(), 51 | nn.MaxPool2d(kernel_size=2) 52 | ) 53 | self.layer2 = nn.Sequential( 54 | nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3, padding=1), 55 | nn.BatchNorm2d(128), 56 | nn.ReLU(), 57 | nn.MaxPool2d(2) 58 | ) 59 | self.layer3 = nn.Sequential( 60 | nn.Conv2d(in_channels=128, out_channels=256, kernel_size=3, padding=1), 61 | nn.BatchNorm2d(256), 62 | nn.ReLU(), 63 | nn.MaxPool2d(2) 64 | ) 65 | self.layer4 = nn.Sequential( 66 | nn.Conv2d(in_channels=256, out_channels=512, kernel_size=3, padding=1), 67 | nn.BatchNorm2d(512), 68 | nn.ReLU(), 69 | nn.MaxPool2d(2) 70 | ) 71 | 72 | # 使用全局平均池化代替Flatten 73 | self.global_avg_pool = nn.AdaptiveAvgPool2d(1) 74 | self.fc = nn.Sequential( 75 | nn.Linear(512, 4096), 76 | nn.Dropout(0.2), 77 | nn.ReLU(), 78 | nn.Linear(4096, gen_ImageCaptcha.captcha_size * len(gen_ImageCaptcha.captcha_array)) 79 | ) 80 | 81 | def forward(self, x): 82 | x = self.layer1(x) 83 | x = self.layer2(x) 84 | x = self.layer3(x) 85 | x = self.layer4(x) 86 | x = self.global_avg_pool(x) # 全局平均池化 87 | x = x.view(x.size(0), -1) # 展平 88 | x = self.fc(x) 89 | return x 90 | ``` 91 | #### 模型训练代码 92 |     这里的代码很常规，分类模型的训练，这里训练24个epoch 93 | ```python 94 | def train(epoch): 95 | train_datas = mydatasets("../dataset/train") 96 | train_dataloader = DataLoader(train_datas, batch_size=64, shuffle=True) 97 | m = MyModel().cuda() 98 | 99 | loss_fn = nn.MultiLabelSoftMarginLoss().cuda() 100 | optimizer = torch.optim.Adam(m.parameters(), lr=0.001) 101 | epoch_losses = [] 102 | for i in range(epoch): 103 | losses = [] 104 | # 迭代器进度条 105 | data_loader_tqdm = tqdm(train_dataloader) 106 | 107 | epoch_loss = 0 108 | for inputs, labels in data_loader_tqdm: 109 | inputs = inputs.cuda() 110 | labels = labels.cuda() 111 | optimizer.zero_grad() 112 | 113 | outputs = m(inputs) 114 | loss = loss_fn(outputs, labels) 115 | losses.append(loss.item()) 116 | epoch_loss = np.mean(losses) 117 | data_loader_tqdm.set_description( 118 | f"This epoch is {str(i + 1)} and it's loss is {loss.item()}, average loss {epoch_loss}" 119 | ) 120 | 121 | loss.backward() 122 | optimizer.step() 123 | epoch_losses.append(epoch_loss) 124 | # 每过一个batch就保存一次模型 125 | torch.save(m.state_dict(), f'../deplo/model/{str(i + 1)}_{epoch_loss}.pth') 126 | 127 | 128 | if __name__ == '__main__': 129 | train(24) 130 | 131 | ``` 132 | loss值变化如下： 133 | ![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/e155e98f50d24c5a9b244c92c9e98956.png#pic_center) 134 | 135 | # 调用 136 | #### 模型预测： 137 |     把模型的输出转换成字符，也就是每隔62就解码一个字符 138 | ```python 139 | def test_pred(): 140 | m = MyModel() 141 | m.load_state_dict(torch.load("../deploy/model/22_0.0007106820558649762.pth")) 142 | m.eval() 143 | test_data = mydatasets("../dataset/test") 144 | 145 | test_dataloader = DataLoader(test_data, batch_size=1, shuffle=False) 146 | test_length = test_data.__len__() 147 | 148 | correct = 0 149 | 150 | for i, (imgs, lables) in enumerate(test_dataloader): 151 | imgs = imgs 152 | lables = lables 153 | lables = lables.view(-1, gen_ImageCaptcha.captcha_array.__len__()) 154 | lables_text = one_hot.vectotext(lables) 155 | start_time = time.time() 156 | predict_outputs = m(imgs) 157 | predict_outputs = predict_outputs.view(-1, gen_ImageCaptcha.captcha_array.__len__()) 158 | predict_labels = one_hot.vectotext(predict_outputs) 159 | print(time.time() - start_time) 160 | if predict_labels == lables_text: 161 | correct += 1 162 | else: 163 | pass 164 | print("正确率{}".format(correct / test_length * 100)) 165 | 166 | 167 | if __name__ == '__main__': 168 | test_pred() 169 | ``` 170 | #### 识别速度和平均成功率如下： 171 | ![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/6c81227c1ecb4f9e81f4eb564d5b3b85.png#pic_center) 172 | 173 | ![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/fcff4e72746c43148ffb16a3b1729f74.png#pic_center) 174 | 175 | ![在这里插入图片描述](https://i-blog.csdnimg.cn/direct/6f4eeb23947542f38c7077563ba9ffe3.jpeg#pic_center) 176 | -------------------------------------------------------------------------------- /development/__pycache__/gen_ImageCaptcha.cpython-38.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/taisuii/ImageCaptchaOCR/a838c3a51355fcefc748205a1ab5976598749ac4/development/__pycache__/gen_ImageCaptcha.cpython-38.pyc -------------------------------------------------------------------------------- /development/__pycache__/model.cpython-38.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/taisuii/ImageCaptchaOCR/a838c3a51355fcefc748205a1ab5976598749ac4/development/__pycache__/model.cpython-38.pyc -------------------------------------------------------------------------------- /development/__pycache__/one_hot.cpython-38.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/taisuii/ImageCaptchaOCR/a838c3a51355fcefc748205a1ab5976598749ac4/development/__pycache__/one_hot.cpython-38.pyc -------------------------------------------------------------------------------- /development/gen_ImageCaptcha.py: -------------------------------------------------------------------------------- 1 | import os 2 | import random 3 | import time 4 | 5 | captcha_array = list("0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ") 6 | captcha_size = 5 7 | from captcha.image import ImageCaptcha 8 | 9 | if __name__ == '__main__': 10 | print(captcha_array) 11 | image = ImageCaptcha() 12 | for i in range(1): 13 | image_val = "".join(random.sample(captcha_array, 4)) 14 | image_name = "./{}_{}.png".format(image_val, int(time.time())) 15 | print(image_name) 16 | image.write(image_val, image_name) 17 | -------------------------------------------------------------------------------- /development/model.py: -------------------------------------------------------------------------------- 1 | import torch 2 | from matplotlib import pyplot as plt 3 | from torch import nn 4 | import development.gen_ImageCaptcha as gen_ImageCaptcha 5 | import os 6 | from PIL import Image 7 | from torch.utils.data import Dataset 8 | from torchvision import transforms 9 | import one_hot 10 | from torch.utils.data import DataLoader 11 | from tqdm import tqdm 12 | import numpy as np 13 | 14 | 15 | class mydatasets(Dataset): 16 | def __init__(self, root_dir): 17 | super(mydatasets, self).__init__() 18 | self.list_image_path = [os.path.join(root_dir, image_name) for image_name in os.listdir(root_dir)] 19 | self.transforms = transforms.Compose([ 20 | transforms.ToTensor() 21 | ]) 22 | 23 | def __getitem__(self, index): 24 | image_path = self.list_image_path[index] 25 | img_ = Image.open(image_path) 26 | image_name = image_path.split("\\")[-1] 27 | img_tesor = self.transforms(img_) 28 | img_lable = image_name.split(".")[0] 29 | img_lable = one_hot.text2vec(img_lable) 30 | img_lable = img_lable.view(1, -1)[0] 31 | return img_tesor, img_lable 32 | 33 | def __len__(self): 34 | return self.list_image_path.__len__() 35 | 36 | 37 | class MyModel(nn.Module): 38 | def __init__(self): 39 | super(MyModel, self).__init__() 40 | self.layer1 = nn.Sequential( 41 | nn.Conv2d(in_channels=3, out_channels=64, kernel_size=3, padding=1), 42 | nn.BatchNorm2d(64), # 添加Batch Normalization 43 | nn.ReLU(), 44 | nn.MaxPool2d(kernel_size=2) 45 | ) 46 | self.layer2 = nn.Sequential( 47 | nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3, padding=1), 48 | nn.BatchNorm2d(128), 49 | nn.ReLU(), 50 | nn.MaxPool2d(2) 51 | ) 52 | self.layer3 = nn.Sequential( 53 | nn.Conv2d(in_channels=128, out_channels=256, kernel_size=3, padding=1), 54 | nn.BatchNorm2d(256), 55 | nn.ReLU(), 56 | nn.MaxPool2d(2) 57 | ) 58 | self.layer4 = nn.Sequential( 59 | nn.Conv2d(in_channels=256, out_channels=512, kernel_size=3, padding=1), 60 | nn.BatchNorm2d(512), 61 | nn.ReLU(), 62 | nn.MaxPool2d(2) 63 | ) 64 | 65 | # 使用全局平均池化代替Flatten 66 | self.global_avg_pool = nn.AdaptiveAvgPool2d(1) 67 | self.fc = nn.Sequential( 68 | nn.Linear(512, 4096), 69 | nn.Dropout(0.2), 70 | nn.ReLU(), 71 | nn.Linear(4096, gen_ImageCaptcha.captcha_size * len(gen_ImageCaptcha.captcha_array)) 72 | ) 73 | 74 | def forward(self, x): 75 | x = self.layer1(x) 76 | x = self.layer2(x) 77 | x = self.layer3(x) 78 | x = self.layer4(x) 79 | x = self.global_avg_pool(x) # 全局平均池化 80 | x = x.view(x.size(0), -1) # 展平 81 | x = self.fc(x) 82 | return x 83 | 84 | 85 | def train(epoch): 86 | train_datas = mydatasets("../dataset/train") 87 | train_dataloader = DataLoader(train_datas, batch_size=64, shuffle=True) 88 | m = MyModel().cuda() 89 | 90 | loss_fn = nn.MultiLabelSoftMarginLoss().cuda() 91 | optimizer = torch.optim.Adam(m.parameters(), lr=0.001) 92 | epoch_losses = [] 93 | for i in range(epoch): 94 | losses = [] 95 | # 迭代器进度条 96 | data_loader_tqdm = tqdm(train_dataloader) 97 | 98 | epoch_loss = 0 99 | for inputs, labels in data_loader_tqdm: 100 | inputs = inputs.cuda() 101 | labels = labels.cuda() 102 | optimizer.zero_grad() 103 | 104 | outputs = m(inputs) 105 | loss = loss_fn(outputs, labels) 106 | losses.append(loss.item()) 107 | epoch_loss = np.mean(losses) 108 | data_loader_tqdm.set_description( 109 | f"This epoch is {str(i + 1)} and it's loss is {loss.item()}, average loss {epoch_loss}" 110 | ) 111 | 112 | loss.backward() 113 | optimizer.step() 114 | epoch_losses.append(epoch_loss) 115 | # 每过一个batch就保存一次模型 116 | torch.save(m.state_dict(), f'../deploy/model/{str(i + 1)}_{epoch_loss}.pth') 117 | 118 | # loss 变化绘制代码 119 | data = np.array(epoch_losses) 120 | plt.figure(figsize=(10, 6)) 121 | plt.plot(data) 122 | plt.title(f"{epoch} epoch loss change") 123 | plt.xlabel("epoch") 124 | plt.ylabel("Loss") 125 | # 显示图像 126 | plt.show() 127 | print(f"completed. Model saved.") 128 | 129 | if __name__ == '__main__': 130 | train(24) 131 | -------------------------------------------------------------------------------- /development/one_hot.py: -------------------------------------------------------------------------------- 1 | import development.gen_ImageCaptcha as gen_ImageCaptcha 2 | import torch 3 | 4 | def text2vec(text): 5 | vectors = torch.zeros((gen_ImageCaptcha.captcha_size, gen_ImageCaptcha.captcha_array.__len__())) 6 | for i in range(len(text)): 7 | vectors[i, gen_ImageCaptcha.captcha_array.index(text[i])] = 1 8 | return vectors 9 | 10 | 11 | def vectotext(vec): 12 | vec = torch.argmax(vec, dim=1) 13 | 14 | text_label = "" 15 | for v in vec: 16 | text_label += gen_ImageCaptcha.captcha_array[v] 17 | return text_label 18 | 19 | 20 | if __name__ == '__main__': 21 | vec = text2vec("aaabv") 22 | print(vec.shape) 23 | print(vectotext(vec)) 24 | -------------------------------------------------------------------------------- /development/predict.py: -------------------------------------------------------------------------------- 1 | import time 2 | 3 | import numpy as np 4 | from PIL import Image 5 | from matplotlib import pyplot as plt 6 | from torch.utils.data import DataLoader 7 | import one_hot 8 | import torch 9 | import development.gen_ImageCaptcha as gen_ImageCaptcha 10 | from torchvision import transforms 11 | from model import mydatasets, MyModel 12 | 13 | 14 | def test_pred(): 15 | m = MyModel() 16 | m.load_state_dict(torch.load("../deploy/model/22_0.0007106820558649762.pth")) 17 | m.eval() 18 | test_data = mydatasets("../dataset/test") 19 | 20 | test_dataloader = DataLoader(test_data, batch_size=1, shuffle=False) 21 | test_length = test_data.__len__() 22 | 23 | correct = 0 24 | spendtime = [] 25 | for i, (imgs, lables) in enumerate(test_dataloader): 26 | imgs = imgs 27 | lables = lables 28 | lables = lables.view(-1, gen_ImageCaptcha.captcha_array.__len__()) 29 | lables_text = one_hot.vectotext(lables) 30 | start_time = time.time() 31 | predict_outputs = m(imgs) 32 | predict_outputs = predict_outputs.view(-1, gen_ImageCaptcha.captcha_array.__len__()) 33 | predict_labels = one_hot.vectotext(predict_outputs) 34 | spendtime.append(time.time() - start_time) 35 | print(predict_labels, lables_text) 36 | if predict_labels == lables_text: 37 | correct += 1 38 | else: 39 | pass 40 | 41 | correct_ = correct / test_length * 100 42 | data = np.array(spendtime) 43 | plt.figure(figsize=(10, 6)) 44 | plt.plot(data) 45 | plt.title(f"verify spend time, average spend: {np.mean(spendtime)} and Success rate: {correct_}") 46 | plt.ylabel("time") 47 | # 显示图像 48 | plt.show() 49 | 50 | 51 | if __name__ == '__main__': 52 | test_pred() 53 | -------------------------------------------------------------------------------- /development/pth2onnx.py: -------------------------------------------------------------------------------- 1 | from model import MyModel 2 | import torch 3 | 4 | def convert(): 5 | # 加载 PyTorch 模型 6 | model_path = "model/resnet18_38_0.021147585306924.pth" 7 | model = MyModel() 8 | model.load_state_dict(torch.load(model_path)) 9 | model.eval() 10 | # 生成一个示例输入 11 | dummy_input = torch.randn(10, 3, 224, 224) 12 | # 将模型转换为 ONNX 格式 13 | torch.onnx.export(model, dummy_input, "model/resnet18.onnx", verbose=True) 14 | 15 | 16 | if __name__ == '__main__': 17 | convert() 18 | -------------------------------------------------------------------------------- /development/server.py: -------------------------------------------------------------------------------- 1 | from flask import Flask, request, jsonify 2 | import torch 3 | from PIL import Image 4 | import io 5 | import base64 6 | import torchvision.transforms as transforms 7 | 8 | from development import gen_ImageCaptcha, one_hot 9 | from development.model import MyModel 10 | 11 | # 加载模型 12 | m = MyModel() 13 | m.load_state_dict(torch.load("../deploy/model/22_0.0007106820558649762.pth")) 14 | m.eval() 15 | 16 | # Flask 应用初始化 17 | app = Flask(__name__) 18 | 19 | # 预处理步骤（如有必要可以自定义） 20 | data_transform = transforms.Compose([ 21 | transforms.ToTensor(), 22 | ]) 23 | 24 | 25 | # OCR 处理函数 26 | def ocr(image_bytes): 27 | # 将字节数据转化为PIL图像 28 | image = Image.open(io.BytesIO(image_bytes)).convert('RGB') 29 | 30 | # 图像预处理 31 | imgs = data_transform(image) 32 | 33 | # 模型推理 34 | predict_outputs = m(imgs.unsqueeze(0)) # 加入批次维度 35 | predict_outputs = predict_outputs.view(-1, gen_ImageCaptcha.captcha_array.__len__()) 36 | 37 | # 转换为可读标签 38 | predict_labels = one_hot.vectotext(predict_outputs) 39 | 40 | return predict_labels 41 | 42 | 43 | # 路由处理 44 | @app.route("/runtime/text/invoke", methods=["POST"]) 45 | def invoke_ocr(): 46 | # 从请求中提取 JSON 47 | req_data = request.get_json() 48 | 49 | # 提取项目名和图像（Base64 编码） 50 | project_name = req_data.get("project_name", "") 51 | image_base64 = req_data.get("image", "") 52 | 53 | if project_name != "ctc_en5l_240516": 54 | return jsonify({"error": "Invalid project name"}), 400 55 | 56 | # 解码 Base64 图像 57 | image_bytes = base64.b64decode(image_base64) 58 | 59 | # 运行 OCR 识别 60 | try: 61 | ocr_result = ocr(image_bytes) 62 | return jsonify({"data": ocr_result}) 63 | except Exception as e: 64 | return jsonify({"error": str(e)}), 500 65 | 66 | 67 | # 启动 Flask 应用 68 | if __name__ == "__main__": 69 | app.run(host="127.0.0.1", port=19199) 70 | -------------------------------------------------------------------------------- /requirement.txt: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/taisuii/ImageCaptchaOCR/a838c3a51355fcefc748205a1ab5976598749ac4/requirement.txt -------------------------------------------------------------------------------- /英数验证码数据集.zip: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/taisuii/ImageCaptchaOCR/a838c3a51355fcefc748205a1ab5976598749ac4/英数验证码数据集.zip --------------------------------------------------------------------------------