基于云边深度学习融合的COVID-19智能检测系统.zip

""" 修改过混淆矩阵后 """ import imageio.v2 as imageio from torch.nn import init from torch.optim.lr_scheduler import CosineAnnealingLR import datetime import itertools import os import random # import cv2 import matplotlib.pyplot as plt import numpy as np # PaddyDataSet import torch import torch.nn as nn import torch.optim as optim import torchvision.transforms as transforms from torchvision import models from PIL import Image from sklearn.metrics import roc_curve, auc from sklearn.model_selection import KFold from torch.utils.data import DataLoader from torch.utils.data import Dataset from torchvision.datasets import ImageFolder # from pytorchtools import EarlyStopping from tqdm import tqdm import GhostNet import efficientnet import GhostNet_res import Covnet import Covnet_2 import Covnet_3 import ResNet # negative = 'cat' # positive = 'dog' # negative = 'negative' # positive = 'positive' # 工作目录 work_path = r"D:\学习\大创\data\训练数据集\model" # 数据集文件夹位置 filepath = r"D:\学习\大创\data\训练数据集\data\Track1+CoughVid 谱图合集\Track1+CoughVid logMel 男女加阴阳" # paddy_labels = {negative: 0, # positive: 1} paddy_labels = {"nf": 0, "nm": 1, "pf": 2, "pm": 3} # 用于存放测试集的数据和标签 test_data_info = [] # 用于包装train和val数据的dataset迭代器，里面剔除了test数据 class PaddyDataSet_train_val(Dataset): def __init__(self, data_dir, transform=None, test_num=0.1): """ 数据集 """ self.label_name = paddy_labels # data_info 存储所有图片路径和标签, 在DataLoader中通过index读取样本 self.data_info = self.get_img_info(data_dir) self.transform = transform self.temp = np.zeros((224, 224)) def __getitem__(self, index): path_img, label = self.data_info[index] img = Image.open(path_img).convert('RGB') # print(img.size) if img.size == self.temp.shape: img = img.resize((224, 224)) # print(img.size) if self.transform is not None: img = self.transform(img) return img, label def __len__(self): return len(self.data_info) @staticmethod def get_img_info(data_dir): data_info = list() for root, dirs, _ in os.walk(data_dir): # 遍历类别 for sub_dir in dirs: img_names = os.listdir(os.path.join(root, sub_dir)) img_names = list(filter(lambda x: x.endswith('.jpg'), img_names)) # 遍历图片 for i in range(len(img_names)): img_name = img_names[i] path_img = os.path.join(root, sub_dir, img_name) # print(sub_dir) label = paddy_labels[sub_dir] data_info.append((path_img, int(label))) # data_info 里面包含了全部的数据以及对应的图片，从中选取 test_num 百分数的数据作为验证集 data_num = len(data_info) # 看一下数据集的长度 data_num data_num = int(test_num * data_num) for i in range(data_num): selected_element = random.choice(data_info) test_data_info.append(selected_element) data_info.remove(selected_element) return data_info # 用于包装test数据的dataset迭代器 class PaddyDataSet_test(Dataset): def __init__(self, transform=None, test_data_info=test_data_info): """ 数据集 """ # data_info 存储所有图片路径和标签, 在DataLoader中通过index读取样本 self.test_info = test_data_info self.transform = transform self.temp = np.zeros((224, 224)) def __getitem__(self, index): path_img, label = self.test_info[index] img = Image.open(path_img).convert('RGB') # print(img.size) if img.size == self.temp.shape: img = img.resize((224, 224)) # print(img.size) if self.transform is not None: img = self.transform(img) return img, label def __len__(self): return len(self.test_info) def init_weights(layer): """ 参数初始化设置使用 :param layer: :return: """ # 如果为卷积层，使用 He initialization 方法正态分布生成值，生成随机数填充张量 if type(layer) == nn.Conv2d: # nn.init.normal_(layer.weight, mean=0, std=1) nn.init.kaiming_normal_(layer.weight, a=0, mode='fan_in', nonlinearity='relu') # 如果为全连接层，权重使用均匀分布初始化，偏置初始化为0.1 elif type(layer) == nn.Linear: nn.init.uniform_(layer.weight, a=-0.1, b=0.1) nn.init.constant_(layer.bias, 0.1) class Focal_Loss(torch.nn.Module): """ 二分类Focal Loss """ def __init__(self, alpha=0.25, gamma=1): super(Focal_Loss, self).__init__() self.alpha = alpha self.gamma = gamma def forward(self, preds, labels): """ preds:sigmoid的输出结果 labels：标签 """ eps = 1e-7 loss = [] for i in range(len(labels)): loss_1 = -1 * (1 - self.alpha) * torch.pow((1 - preds[i][1]), self.gamma) * torch.log(preds[i][1] + eps) * \ labels[i] loss_0 = -1 * self.alpha * torch.pow(preds[i][1], self.gamma) * torch.log(1 - preds[i][1] + eps) * ( 1 - labels[i]) loss.append(loss_0 + loss_1) long = len(loss) loss = sum(loss) loss = torch.as_tensor(loss / long) # return torch.mean(loss) return loss def getStat(all_data): ''' 用于计算自己（图片）数据集的均值与方差 :param train_data: 自定义类Dataset(或ImageFolder即可) :return: (mean, std) ''' train_loader = torch.utils.data.DataLoader( all_data, batch_size=1, shuffle=False, num_workers=0, pin_memory=True) mean = torch.zeros(3) std = torch.zeros(3) print(type(train_loader)) print(len(all_data)) all_num = len(all_data) num = 0 for X, _ in train_loader: num += 1 print("共{}个，第{}个".format(all_num, num)) for d in range(3): mean[d] += X[:, d, :, :].mean() std[d] += X[:, d, :, :].std() mean.div_(len(all_data)) std.div_(len(all_data)) return list(mean.numpy()), list(std.numpy()) # -------------------------------------------------- # # （0）参数设置 # -------------------------------------------------- # batch_size = 8 # 每个step训练batch_size张图片 epochs = 128 # 共训练epochs次 k = 5 # k折交叉验证 dropout_num_1 = 0.7 dropout_num_2 = 0.7 resnet_dropout = 0.4 learning_rate = 1e-4 pre_score_k = [] labels_k = [] # wd：正则化惩罚的参数 wd = 0.6 print("wd:{}".format(wd)) # wd = None # stop_epoch: 早停的批量数 stop_epoch = 4 # -------------------------------------------------- # # （1）文件配置 # -------------------------------------------------- # # 计算图片的总数量 all_photo_num = 0 for root, dir_list, files in os.walk(filepath): for dir in dir_list: all_photo_num += len(os.listdir(os.path.join(root, dir))) print(all_photo_num) # test_num: 验证集占总数据的比例 test_num = 0.1 test_file_num = (int(test_num * all_photo_num) // batch_size) * batch_size print(test_file_num) # 需要用到train_num 初始化一下混淆矩阵 train_num = all_photo_num * (1 - test_num) * 0.8 # 显示一下文件夹的名称 dir_path = os.path.basename(filepath) print(dir_path) # 创建权重的文件夹 savepath = os.path.join(work_path, 'pth', dir_path) cd = os.path.exists(savepath) if cd: print("权重保存文件夹已存在") else: print("创建权重保存文件夹") os.mkdir(savepath) # 判断保存图�