基于卷积神经网络的乳腺癌细胞识别系统，resnet50，mobilenet模型【pytorch框架+python源码】

    def main(self, epochs):# 记录训练过程log_file_name = './results/resnet50训练和验证过程.txt'# 记录正常的 print 信息sys.stdout = Logger(log_file_name)print("using {} device.".format(self.device))# 开始训练，记录开始时间begin_time = time()# 加载数据train_loader, validate_loader, class_names, train_num, val_num = self.data_load()print("class_names: ", class_names)train_steps = len(train_loader)val_steps = len(validate_loader)# 加载模型model = self.model_load()  # 创建模型# 修改全连接层的输出维度in_channel = model.fc.in_featuresmodel.fc = nn.Linear(in_channel, len(class_names))# 模型结构可视化x = torch.randn(16, 3, 224, 224)  # 随机生成一个输入# 模型结构保存路径model_visual_path = 'results/resnet50_visual.onnx'# 将 pytorch 模型以 onnx 格式导出并保存torch.onnx.export(model, x, model_visual_path)  # netron.start(model_visual_path)  # 浏览器会自动打开网络结构# 将模型放入GPU中model.to(self.device)# 定义损失函数loss_function = nn.CrossEntropyLoss()# 定义优化器params = [p for p in model.parameters() if p.requires_grad]optimizer = optim.Adam(params=params, lr=0.0001)train_loss_history, train_acc_history = [], []test_loss_history, test_acc_history = [], []best_acc = 0.0for epoch in range(0, epochs):# 下面是模型训练model.train()running_loss = 0.0train_acc = 0.0train_bar = tqdm(train_loader, file=sys.stdout)# 进来一个batch的数据，计算一次梯度，更新一次网络for step, data in enumerate(train_bar):# 获取图像及对应的真实标签images, labels = data# 清空过往梯度optimizer.zero_grad()# 得到预测的标签outputs = model(images.to(self.device))# 计算损失train_loss = loss_function(outputs, labels.to(self.device))# 反向传播，计算当前梯度train_loss.backward()# 根据梯度更新网络参数optimizer.step()  # 累加损失running_loss += train_loss.item()# 每行最大值的索引predict_y = torch.max(outputs, dim=1)[1]  # torch.eq()进行逐元素的比较，若相同位置的两个元素相同，则返回True；若不同，返回Falsetrain_acc += torch.eq(predict_y, labels.to(self.device)).sum().item()# 更新进度条train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(epoch + 1,epochs,train_loss)# 下面是模型验证# 不启用 BatchNormalization 和 Dropout，保证BN和dropout不发生变化model.eval()# accumulate accurate number / epochval_acc = 0.0  testing_loss = 0.0# 张量的计算过程中无需计算梯度with torch.no_grad():  val_bar = tqdm(validate_loader, file=sys.stdout)for val_data in val_bar:# 获取图像及对应的真实标签val_images, val_labels = val_data# 得到预测的标签outputs = model(val_images.to(self.device))# 计算损失val_loss = loss_function(outputs, val_labels.to(self.device))  testing_loss += val_loss.item()# 每行最大值的索引predict_y = torch.max(outputs, dim=1)[1]  # torch.eq()进行逐元素的比较，若相同位置的两个元素相同，则返回True；若不同，返回Falseval_acc += torch.eq(predict_y, val_labels.to(self.device)).sum().item()train_loss = running_loss / train_stepstrain_accurate = train_acc / train_numtest_loss = testing_loss / val_stepsval_accurate = val_acc / val_numtrain_loss_history.append(train_loss)train_acc_history.append(train_accurate)test_loss_history.append(test_loss)test_acc_history.append(val_accurate)print('[epoch %d] train_loss: %.3f  val_accuracy: %.3f' %(epoch + 1, train_loss, val_accurate))# 保存最佳模型if val_accurate > best_acc:best_acc = val_accuratetorch.save(model.state_dict(), self.model_name)# 记录结束时间end_time = time()run_time = end_time - begin_timeprint('该循环程序运行时间：', run_time, "s")# 绘制模型训练过程图self.show_loss_acc(train_loss_history, train_acc_history,test_loss_history, test_acc_history)# 画热力图test_real_labels, test_pre_labels = self.heatmaps(model, validate_loader, class_names)# 计算混淆矩阵self.calculate_confusion_matrix(test_real_labels, test_pre_labels, class_names)