赞
踩
基于卷积神经网络的花卉识别系统是在pytorch框架下实现的,系统中有两个模型可选resnet50模型和VGG16模型,这两个模型可用于模型效果对比。该系统涉及的技术栈有,UI界面:python + pyqt5,前端界面:python + flask
该项目是在pycharm和anaconda搭建的虚拟环境执行,pycharm和anaconda安装和配置可观看教程:
超详细的pycharm+anaconda搭建python虚拟环境_pycharm配置anaconda虚拟环境-CSDN博客
pycharm+anaconda搭建python虚拟环境_哔哩哔哩_bilibili
- class MainProcess:
- def __init__(self, train_path, test_path, model_name):
- self.train_path = train_path
- self.test_path = test_path
- self.model_name = model_name
- self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-
- def main(self, epochs):
- # 记录训练过程
- log_file_name = './results/vgg16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
-
- # 网络结构可视化
- x = torch.randn(16, 3, 224, 224) # 随机生成一个输入
- model_visual_path = 'results/vgg16_visual.onnx' # 模型结构保存路径
- torch.onnx.export(model, x, model_visual_path) # 将 pytorch 模型以 onnx 格式导出并保存
- # netron.start(model_visual_path) # 浏览器会自动打开网络结构
-
- # load pretrain weights
- # download url: https://download.pytorch.org/models/vgg16-397923af.pth
- model_weight_path = "models/vgg16-pre.pth"
- assert os.path.exists(model_weight_path), "file {} does not exist.".format(model_weight_path)
- model.load_state_dict(torch.load(model_weight_path, map_location='cpu'))
-
- # 更改Vgg16模型的最后一层
- model.classifier[-1] = nn.Linear(4096, len(class_names), bias=True)
-
- # 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 = 0.0
-
- for epoch in range(0, epochs):
- # 下面是模型训练
- model.train()
- running_loss = 0.0
- train_acc = 0.0
- train_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):
- images, labels = data # 获取图像及对应的真实标签
- optimizer.zero_grad() # 清空过往梯度
- outputs = model(images.to(self.device)) # 得到预测的标签
- train_loss = loss_function(outputs, labels.to(self.device)) # 计算损失
- train_loss.backward() # 反向传播,计算当前梯度
- optimizer.step() # 根据梯度更新网络参数
-
- # print statistics
- running_loss += train_loss.item()
- predict_y = torch.max(outputs, dim=1)[1] # 每行最大值的索引
- # 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False
- train_acc += torch.eq(predict_y, labels.to(self.device)).sum().item()
- train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(epoch + 1,
- epochs,
- train_loss)
- # 下面是模型验证
- model.eval() # 不启用 BatchNormalization 和 Dropout,保证BN和dropout不发生变化
- val_acc = 0.0 # accumulate accurate number / epoch
- testing_loss = 0.0
- with torch.no_grad(): # 张量的计算过程中无需计算梯度
- val_bar = tqdm(validate_loader, file=sys.stdout)
- for val_data in val_bar:
- val_images, val_labels = val_data
- outputs = model(val_images.to(self.device))
-
- val_loss = loss_function(outputs, val_labels.to(self.device)) # 计算损失
- testing_loss += val_loss.item()
-
- predict_y = torch.max(outputs, dim=1)[1] # 每行最大值的索引
- # 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False
- val_acc += torch.eq(predict_y, val_labels.to(self.device)).sum().item()
-
- train_loss = running_loss / train_steps
- train_accurate = train_acc / train_num
- test_loss = testing_loss / val_steps
- val_accurate = val_acc / val_num
-
- train_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))
- if val_accurate > best_acc:
- best_acc = val_accurate
- torch.save(model.state_dict(), self.model_name)
-
- # 记录结束时间
- end_time = time()
- run_time = end_time - begin_time
- print('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
- self.heatmaps(model, validate_loader, class_names)
该系统可以训练自己的数据集,训练过程也比较简单,只需指定自己数据集中训练集和测试集的路径,训练后模型名称和指定训练的轮数即可
编码不易,源码有偿获取喔!
资源主要包括以下内容:完整的程序代码文件、训练好的模型、数据集、UI界面、前端界面。欢迎大家咨询!
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。