Python基于改进ResNet50的结核病基因耐药检测系统源代码

import argparse import json import shutil from django.utils import timezone import torch.nn.functional as F import numpy as np import torch from torchvision.models import resnet50 from django.http import HttpResponse from django.shortcuts import render, redirect from django.views import View from django.conf import settings from django.core.files.storage import FileSystemStorage import os from detection.cts_predict import svm_predict, load_config, load_data, load_model from detection.getTestInfo import getFileImgsNum, flattenList, classify_Img_byPatient, deal_patient_result, print_patientInfo, print_fault_patientInfo from detection.model import ResNet50 from detection.resnet50_predict import resnet50_main log_file = '' resistance = '0' # 上传测试集，接收测试集 class Upload(View): """预测功能：上传测试集""" def get(self, request): img_path = os.path.join(settings.MEDIA_ROOT, 'test') if os.listdir(img_path): # 检测之前情况上一次检测的残留图片 for f in os.listdir(img_path): file_p = os.path.join(img_path, f) try: if os.path.isdir(file_p): shutil.rmtree(file_p) except: pass return render(request, 'detection/detection-upload.html', context={'message': "上传数据"}) def post(self, request): try: global resistance # 将文件夹中的每一张图片都写入 resistance = request.POST.get('resistance') # 是否耐药 folder_name = request.POST.get('folder_name') # 患者姓名设置为文件夹名称 if resistance == '1': fs = FileSystemStorage(location=settings.MEDIA_ROOT + '/test/Rifampicin-intolerant/' + folder_name) elif resistance == '0': fs = FileSystemStorage(location=settings.MEDIA_ROOT + '/test/Nelrifampicin/' + folder_name) files = request.FILES.getlist('files') for file in files: fs.save(file.name, file) return redirect('detection:execute_detection') except Exception as e: print(e) return render(request, 'detection/detection-upload.html', context={'message': '上传出错，请重新上传'}) def execute_detection(request): global log_file # 创建日志文件 log_file_path = os.path.join(settings.STATICFILES_DIRS[0], 'predict_res') if not os.path.exists(log_file_path): os.makedirs(log_file_path) log_file = log_file_path + '/predict.txt' if not os.path.exists(log_file): with open(log_file, 'a') as f: f.write('[]') # 获取参数 cst_args = load_config() # 导入模型 resnet50_model = load_resnet_model() # 加载数据集 test_dataset, test_loader = load_data(cst_args) # 类别个数 num_classes = 2 # 获取当前数据集的图片大小维度 data, target = next(iter(test_loader)) # 初始化cst_model的参数 resolution = (data.shape[2], data.shape[3]) # 定义创建cst模型 cst_model = ResNet50(num_classes=num_classes, resolution=resolution, heads=4) # 加载预训练权重 cst_model, svm_model = load_model(cst_args, cst_model) # 获取数据集信息 file_datas = getFileImgsNum(test_dataset, os.path.join(cst_args.get('data_path'), 'test')) predictions, labels, infos = predict(test_loader, resnet50_model, cst_model, svm_model, cst_args, file_datas) # 每张图的预测结果统计 num1 = 0 num0 = 0 acc = 0 correct_count = int(request.COOKIES.get('correct')) error_count = int(request.COOKIES.get('error')) for cont in predictions[0]: if cont: num1 += 1 else: num0 += 1 if resistance == '1': print(f'患者{infos}应该是不耐药的') result = '耐药' if num0 >= num1 else '不耐药' if result == '不耐药': correct_count += 1 acc = 1 print('预测正确') else: error_count += 1 acc = 0 print('预测错误') elif resistance == '0': print(f'患者{infos}应该是耐药的') result = '耐药' if num0 >= num1 else '不耐药' if result == '耐药': correct_count += 1 acc = 1 print('预测正确') else: error_count += 1 acc = 0 print('预测错误') # 返回预测结果到结果可视化界面 context = { 'predictions': predictions[0], 'labels': labels[0], 'x_label': [num0, num1], 'infos': infos, 'max': len(predictions[0]), 'result': '耐药' if num0 >= num1 else '不耐药', 'acc': '预测正确' if acc else '预测错误', } with open(log_file, 'r') as f: text = f.read() print('text', text) json_data = json.loads(text) print(json_data) json_data.append({ "name": infos, "acc": acc, "datetime": timezone.now().strftime("%Y-%m-%d %H:%M:%S"), }) with open(log_file, 'w') as f: f.write(json.dumps(json_data)) response = render(request, 'detection/detection-visualization.html', context=context) predict_count = int(request.COOKIES.get('predict')) predict_count += 1 response.set_cookie('predict', predict_count) response.set_cookie('correct', correct_count) response.set_cookie('error', error_count) return response def predict(test_loader, resnet50_model, cst_model, svm_model, args, file_datas): global log_file resnet50_model.eval() cst_model.eval() acc = 0. predicted_result = [] labels_result = [] # svm_mode = SVM_model(kernel='rbf', C=1.0, class_weight ='balanced') with torch.no_grad(): for data, target in test_loader: if args.get('cuda'): data, target = data.cuda(), target.cuda() output_resnet50 = resnet50_model(data) output_cst = cst_model(data) # 权重效果 # cst:275 # 0.1:0.9 Test Acc: 80.352 # 0.2:0.8 Test Acc: 81.496 # 0.3:0.7 Test Acc: 82.459 # 0.4:0.6 Test Acc: 83.112 # 0.5:0.5 Test Acc: 83.494 # 0.6:0.4 Test Acc: 83.657 # 0.7:0.3 Test Acc: 83.893 效果最好 # 0.8:0.2 Test Acc: 83.766 # 0.9:0.1 Test Acc: 83.131 resnet_b = 0.7 cts_b = 0.3 output = resnet_b*output_resnet50 + cts_b*output_cst # SVM---------------------------------- # y_pred = svm_predict(output, data, target, svm_model, isTrain=False) # target_numpy = target.cpu().numpy() # ac = svm_mode.accuracy(target_numpy, y_pred) # acc_current = np.count_nonzero(y_pred == target_numpy) # acc += acc_current # SVM---------------------------------- _, y_pred = F.softmax(output, dim=-1).max(1) acc += y_pred.eq(target).sum().item() # acc += acc_current # 将predicted和labels转换为列表，并添加到对应的子列表中 predicted_result.append(y_pred.tolist()) labels_result.append(target.tolist()) # 将predicted_result中的多个子列表展开成一个平面列表 predicted_result, labels_result = flattenList(labels_result, predicted_result) # 重新对结果进行分类，按照病人进行分类 predicted_result, labels_result = classify_Img_byPatient(file_datas, labels_result, predicted_result) # 输出结果 # 最后的参数是阈值 correct_patient_labels, total_patient_labels, \ Filter_unqualified_data_infos, Filter_unqualified_data_Predictions,\ Filter_unqualified_data_Labels, Fil