深度学习 Day29——利用Pytorch实现咖啡豆识别

一、前言

🍨 本文为🔗365天深度学习训练营 中的学习记录博客

🍦 参考文章：Pytorch实战 | 第P7周：咖啡豆识别(训练营内部成员可读)

🍖 原作者：K同学啊|接辅导、项目定制

本期博客我们将探索完成使用Pytorch框架搭建VGG16网络模型进行咖啡豆的识别任务。

二、我的环境

• 电脑系统：Windows 11
• 语言环境：Python 3.8.5
• 编译器：DataSpell
• 深度学习环境：
  • torch 1.12.1+cu113
  • torchvision 0.13.1+cu113
• 显卡及显存：RTX 3070 8G

三、前期工作

1、导入依赖项设置GPU

import torch import torch.nn as nn import torchvision.transforms as transforms import torchvision from torchvision import transforms, datasets import os,PIL,pathlib,warnings warnings.filterwarnings("ignore") #忽略警告信息  device = torch.device("cuda" if torch.cuda.is_available() else "cpu") device 

device(type='cuda') 

2、导入数据

import os,PIL,random,pathlib data_dir = 'E:\深度学习\data\Day16' data_dir = pathlib.Path(data_dir) data_paths = list(data_dir.glob('*')) classeNames = [str(path).split("\\")[4] for path in data_paths] classeNames 

['Dark', 'Green', 'Light', 'Medium'] 

使用transforms.Compose对数据进行预处理的方法，包括将输入图片resize成统一尺寸、归一化处理等。其中，train_transforms和test_transform是训练集和测试集的预处理方法，total_data是处理后的数据集。

train_transforms = transforms.Compose([ transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸 # transforms.RandomHorizontalFlip(), # 随机水平翻转 transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间 transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛 mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。 ]) test_transform = transforms.Compose([ transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸 transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间 transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛 mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。 ]) total_data = datasets.ImageFolder(data_dir,transform=train_transforms) total_data 

Dataset ImageFolder Number of datapoints: 1200 Root location: E:\深度学习\data\Day16 StandardTransform Transform: Compose( Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=None) ToTensor() Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) ) 

total_data.class_to_idx # 查看类别对应的索引 

{ 
   'Dark': 0, 'Green': 1, 'Light': 2, 'Medium': 3} 

3、划分数据集

将数据集分为训练集和测试集，首先通过数据总量的80%来计算训练集大小，然后用总量减去训练集大小得到测试集大小，最后使用PyTorch的random_split函数将数据集随机分为训练集和测试集。

train_size = int(0.8 * len(total_data)) test_size = len(total_data) - train_size train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size]) train_dataset, test_dataset 

(<torch.utils.data.dataset.Subset at 0x2168b>, <torch.utils.data.dataset.Subset at 0x2168b369df0>) 

定义两个数据加载器，分别是train_dl和test_dl。每个加载器都有一个batch_size参数，用于指定每个批次的大小。此外，还有shuffle和num_workers参数，用于打乱数据集并指定使用的线程数。

batch_size = 32 train_dl = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True,num_workers=1) test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle=True,num_workers=1) 

使用test_dl数据集进行迭代，输出X和y的形状和数据类型。其中X的形状为[N, C, H, W]，y的形状和数据类型则分别输出。

for x, y in test_dl: print(x.shape, y.shape) break 

torch.Size([32, 3, 224, 224]) torch.Size([32]) 

四、手动搭建VGG16模型

1、模型搭建

import torch.nn.functional as F class vgg16(nn.Module): def __init__(self): super(vgg16, self).__init__() # 卷积块1 self.block1 = nn.Sequential( nn.Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)) ) # 卷积块2 self.block2 = nn.Sequential( nn.Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)) ) # 卷积块3 self.block3 = nn.Sequential( nn.Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)) ) # 卷积块4 self.block4 = nn.Sequential( nn.Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)) ) # 卷积块5 self.block5 = nn.Sequential( nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), nn.ReLU(), nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2)) ) # 全连接网络层，用于分类 self.classifier = nn.Sequential( nn.Linear(in_features=512*7*7, out_features=4096), nn.ReLU(), nn.Linear(in_features=4096, out_features=4096), nn.ReLU(), nn.Linear(in_features=4096, out_features=4) ) def forward(self, x): x = self.block1(x) x = self.block2(x) x = self.block3(x) x = self.block4(x) x = self.block5(x) x = torch.flatten(x, start_dim=1) x = self.classifier(x) return x device = "cuda" if torch.cuda.is_available() else "cpu" print("Using {} device".format(device)) model = vgg16().to(device) model 

Using cuda device Output exceeds the size limit. Open the full output data in a text editor vgg16( (block1): Sequential( (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (1): ReLU() (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (3): ReLU() (4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False) ) (block2): Sequential( (0): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (1): ReLU() (2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (3): ReLU() (4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False) ) (block3): Sequential( (0): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (1): ReLU() (2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (3): ReLU() (4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (5): ReLU() (6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False) ) (block4): Sequential( ... (2): Linear(in_features=4096, out_features=4096, bias=True) (3): ReLU() (4): Linear(in_features=4096, out_features=4, bias=True) ) ) 

这是一个使用PyTorch实现的VGG16模型，包括5个卷积块和1个全连接网络层，用于分类。模型输入为3通道的图像，输出为4个类别的概率分布。模型使用ReLU作为激活函数，最大池化作为下采样方式。模型已移植到GPU上。

2、查看模型参数

# 查看模型参数 for name, param in model.named_parameters(): print(name, '\t', param.shape) 

block1.0.weight torch.Size([64, 3, 3, 3]) block1.0.bias torch.Size([64]) block1.2.weight torch.Size([64, 64, 3, 3]) block1.2.bias torch.Size([64]) block2.0.weight torch.Size([128, 64, 3, 3]) block2.0.bias torch.Size([128]) block2.2.weight torch.Size([128, 128, 3, 3]) block2.2.bias torch.Size([128]) block3.0.weight torch.Size([256, 128, 3, 3]) block3.0.bias torch.Size([256]) block3.2.weight torch.Size([256, 256, 3, 3]) block3.2.bias torch.Size([256]) block3.4.weight torch.Size([256, 256, 3, 3]) block3.4.bias torch.Size([256]) block4.0.weight torch.Size([512, 256, 3, 3]) block4.0.bias torch.Size([512]) block4.2.weight torch.Size([512, 512, 3, 3]) block4.2.bias torch.Size([512]) block4.4.weight torch.Size([512, 512, 3, 3]) block4.4.bias torch.Size([512]) block5.0.weight torch.Size([512, 512, 3, 3]) block5.0.bias torch.Size([512]) block5.2.weight torch.Size([512, 512, 3, 3]) block5.2.bias torch.Size([512]) block5.4.weight torch.Size([512, 512, 3, 3]) ... classifier.2.weight torch.Size([4096, 4096]) classifier.2.bias torch.Size([4096]) classifier.4.weight torch.Size([4, 4096]) classifier.4.bias torch.Size([4]) 

3、调用官方的VGG16网络框架

可以使用 PyTorch 提供的 torchvision.models 模块中的 vgg16 函数来调用官方的 VGG16 网络框架。

import torch import torch.nn as nn import torchvision.models as models # 加载预训练的 VGG16 模型 model = models.vgg16(pretrained=True) # 将模型移动到 GPU 上 device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model.to(device) # 定义损失函数和优化器 criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=0.001) # 训练模型 for epoch in range(num_epochs): for images, labels in train_loader: images = images.to(device) labels = labels.to(device) optimizer.zero_grad() outputs = model(images) loss = criterion(outputs, labels) loss.backward() optimizer.step() # 测试模型 with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Accuracy: {} %'.format(100 * correct / total)) 

我们首先加载了预训练的 VGG16 模型，然后将模型移动到 GPU 上，并定义了损失函数和优化器。在训练过程中，我们遍历训练集，计算损失并反向传播更新参数。在测试过程中，我们遍历测试集，计算模型的准确率。

五、训练模型

1、训练函数

# 训练循环 def train(dataloader, model, loss_fn, optimizer): size = len(dataloader.dataset) # 训练集的大小 num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整) train_loss, train_acc = 0, 0 # 初始化训练损失和正确率 for X, y in dataloader: # 获取图片及其标签 X, y = X.to(device), y.to(device) # 计算预测误差 pred = model(X) # 网络输出 loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失 # 反向传播 optimizer.zero_grad() # grad属性归零 loss.backward() # 反向传播 optimizer.step() # 每一步自动更新 # 记录acc与loss train_acc += (pred.argmax(1) == y).type(torch.float).sum().item() train_loss += loss.item() train_acc /= size train_loss /= num_batches return train_acc, train_loss 

该函数为训练循环，输入数据集、模型、损失函数和优化器，返回训练损失和正确率。函数中进行了模型预测、损失计算、反向传播和参数更新等操作。最终计算并返回训练集的正确率和损失。

2、测试函数

def test (dataloader, model, loss_fn): size = len(dataloader.dataset) # 测试集的大小 num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整) test_loss, test_acc = 0, 0 # 当不进行训练时，停止梯度更新，节省计算内存消耗 with torch.no_grad(): for imgs, target in dataloader: imgs, target = imgs.to(device), target.to(device) # 计算loss target_pred = model(imgs) loss = loss_fn(target_pred, target) test_loss += loss.item() test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item() test_acc /= size test_loss /= num_batches return test_acc, test_loss 

该函数用于测试模型在给定数据集上的准确率和损失。函数会遍历数据集并计算每个批次的损失和准确率，最后返回整个数据集的平均准确率和平均损失。

3、正式训练

import copy optimizer = torch.optim.Adam(model.parameters(), lr= 1e-4) loss_fn = nn.CrossEntropyLoss() # 创建损失函数 epochs = 40 train_loss = [] train_acc = [] test_loss = [] test_acc = [] best_acc = 0 # 设置一个最佳准确率，作为最佳模型的判别指标 for epoch in range(epochs): model.train() epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer) model.eval() epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn) # 保存最佳模型到 best_model if epoch_test_acc > best_acc: best_acc = epoch_test_acc best_model = copy.deepcopy(model) train_acc.append(epoch_train_acc) train_loss.append(epoch_train_loss) test_acc.append(epoch_test_acc) test_loss.append(epoch_test_loss) # 获取当前的学习率 lr = optimizer.state_dict()['param_groups'][0]['lr'] template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}') print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss, lr)) # 保存最佳模型到文件中 PATH = './best_model.pth' # 保存的参数文件名 torch.save(model.state_dict(), PATH) print('Done') 

Epoch: 1, Train_acc:25.8%, Train_loss:1.378, Test_acc:21.2%, Test_loss:1.311, Lr:1.00E-04 Epoch: 2, Train_acc:52.6%, Train_loss:1.062, Test_acc:58.3%, Test_loss:0.798, Lr:1.00E-04 Epoch: 3, Train_acc:59.4%, Train_loss:0.829, Test_acc:71.2%, Test_loss:0.765, Lr:1.00E-04 Epoch: 4, Train_acc:65.8%, Train_loss:0.695, Test_acc:71.7%, Test_loss:0.652, Lr:1.00E-04 Epoch: 5, Train_acc:74.7%, Train_loss:0.576, Test_acc:66.7%, Test_loss:0.577, Lr:1.00E-04 Epoch: 6, Train_acc:74.7%, Train_loss:0.549, Test_acc:77.5%, Test_loss:0.539, Lr:1.00E-04 Epoch: 7, Train_acc:81.4%, Train_loss:0.425, Test_acc:82.5%, Test_loss:0.370, Lr:1.00E-04 Epoch: 8, Train_acc:82.3%, Train_loss:0.381, Test_acc:85.4%, Test_loss:0.315, Lr:1.00E-04 Epoch: 9, Train_acc:84.8%, Train_loss:0.336, Test_acc:85.8%, Test_loss:0.320, Lr:1.00E-04 Epoch:10, Train_acc:87.8%, Train_loss:0.303, Test_acc:91.7%, Test_loss:0.205, Lr:1.00E-04 Epoch:11, Train_acc:90.4%, Train_loss:0.247, Test_acc:89.6%, Test_loss:0.349, Lr:1.00E-04 Epoch:12, Train_acc:94.2%, Train_loss:0.163, Test_acc:97.1%, Test_loss:0.062, Lr:1.00E-04 Epoch:13, Train_acc:96.1%, Train_loss:0.139, Test_acc:86.2%, Test_loss:0.367, Lr:1.00E-04 Epoch:14, Train_acc:95.0%, Train_loss:0.154, Test_acc:93.3%, Test_loss:0.203, Lr:1.00E-04 Epoch:15, Train_acc:97.5%, Train_loss:0.066, Test_acc:98.3%, Test_loss:0.027, Lr:1.00E-04 Epoch:16, Train_acc:96.9%, Train_loss:0.072, Test_acc:97.9%, Test_loss:0.079, Lr:1.00E-04 Epoch:17, Train_acc:98.0%, Train_loss:0.055, Test_acc:98.3%, Test_loss:0.032, Lr:1.00E-04 Epoch:18, Train_acc:98.5%, Train_loss:0.031, Test_acc:100.0%, Test_loss:0.008, Lr:1.00E-04 Epoch:19, Train_acc:99.0%, Train_loss:0.025, Test_acc:99.6%, Test_loss:0.021, Lr:1.00E-04 Epoch:20, Train_acc:98.0%, Train_loss:0.065, Test_acc:98.3%, Test_loss:0.045, Lr:1.00E-04 Epoch:21, Train_acc:96.6%, Train_loss:0.099, Test_acc:96.2%, Test_loss:0.112, Lr:1.00E-04 Epoch:22, Train_acc:99.5%, Train_loss:0.025, Test_acc:99.2%, Test_loss:0.015, Lr:1.00E-04 Epoch:23, Train_acc:99.8%, Train_loss:0.006, Test_acc:99.6%, Test_loss:0.012, Lr:1.00E-04 Epoch:24, Train_acc:95.2%, Train_loss:0.156, Test_acc:96.7%, Test_loss:0.114, Lr:1.00E-04 Epoch:25, Train_acc:98.9%, Train_loss:0.039, Test_acc:98.8%, Test_loss:0.033, Lr:1.00E-04 ... Epoch:38, Train_acc:99.8%, Train_loss:0.005, Test_acc:99.6%, Test_loss:0.016, Lr:1.00E-04 Epoch:39, Train_acc:100.0%, Train_loss:0.000, Test_acc:99.6%, Test_loss:0.007, Lr:1.00E-04 Epoch:40, Train_acc:100.0%, Train_loss:0.000, Test_acc:99.6%, Test_loss:0.006, Lr:1.00E-04 Done 

六、Loss-Accuracy图可视化

import matplotlib.pyplot as plt #隐藏警告 import warnings warnings.filterwarnings("ignore") #忽略警告信息 plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签 plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号 plt.rcParams['figure.dpi'] = 100 #分辨率 epochs_range = range(epochs) plt.figure(figsize=(12, 3)) plt.subplot(1, 2, 1) plt.plot(epochs_range, train_acc, label='Training Accuracy') plt.plot(epochs_range, test_acc, label='Test Accuracy') plt.legend(loc='lower right') plt.title('Training and Validation Accuracy') plt.subplot(1, 2, 2) plt.plot(epochs_range, train_loss, label='Training Loss') plt.plot(epochs_range, test_loss, label='Test Loss') plt.legend(loc='upper right') plt.title('Training and Validation Loss') plt.show() 

七、预测

from PIL import Image classes = list(total_data.class_to_idx) def predict_one_image(image_path, model, transform, classes): test_img = Image.open(image_path).convert('RGB') plt.imshow(test_img) # 展示预测的图片 test_img = transform(test_img) img = test_img.to(device).unsqueeze(0) model.eval() output = model(img) _,pred = torch.max(output,1) pred_class = classes[pred] print(f'预测结果是：{ 
     pred_class}') # 预测训练集中的某张照片 predict_one_image(image_path='E:\深度学习\data\Day16\Medium\medium (1).png', model=model, transform=train_transforms, classes=classes) 

八、模型评估

best_model.eval() epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn) epoch_test_acc, epoch_test_loss 

(1.0, 0.0082927) 

我这里训练之后准确率达到了100%。

九、使用官方MobileNetV2 模型

import torch import torch.nn as nn import torchvision.models as models num_epochs = 20 # 加载预训练的 MobileNetV2 模型 model = models.mobilenet_v2(pretrained=True) # 将最后一层替换为新的全连接层 model.classifier = nn.Sequential( nn.Dropout(0.2), nn.Linear(1280, 4), ) # 将模型移动到 GPU 上 device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model.to(device) # 定义损失函数和优化器 criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=0.001) # 训练模型 for epoch in range(num_epochs): for images, labels in train_dl: images = images.to(device) labels = labels.to(device) optimizer.zero_grad() outputs = model(images) loss = criterion(outputs, labels) loss.backward() optimizer.step() # 测试模型 with torch.no_grad(): correct = 0 total = 0 for images, labels in test_dl: images = images.to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Accuracy: {} %'.format(100 * correct / total)) 

MobileNetV2 模型和 VGG16 模型在网络结构上有很大的不同。MobileNetV2 模型采用了深度可分离卷积的设计，可以在保持较高准确率的同时，大幅度减小模型的参数量和计算量，因此更加轻量级。而 VGG16 模型则是传统的卷积神经网络，参数量和计算量相对较大。