深度学习框架目标检测算法使用YOLOv8进行训练电力设备航拍类/配网缺陷检测无人机航拍图像数据集的代码。进行模型训练和评估。文章目录深度学习框架目标检测算法使用YOLOv8进行训练电力设备航拍类/配网缺陷检测无人机航拍图像数据集的代码。进行模型训练和评估。数据集介绍数据集概述数据集目录结构数据集配置文件转换标注格式转换脚本YOLOv8训练代码详细解释运行训练脚本评估模型详细解释运行评估脚本一键运行脚本详细解释运行主脚本转换标注格式训练模型评估模型数据集介绍数据集概述数据集名称Power Distribution Network Defect Detection Dataset (PDNDDD)数据类型无人机航拍图像目标类别2类不规范绑扎螺栓销钉缺失样本数量约3000张图片标注格式VOC格式数据集目录结构PDNDDD/ ├── images/ │ ├── train/ │ └── val/ ├── labels_voc/ │ ├── train/ │ └── val/ ├── labels/ │ ├── train/ │ └── val/ └── data.yaml数据集配置文件创建一个data.yaml文件配置数据集的路径和类别信息path:./PDNDDD# 数据集路径train:images/train# 训练集图像路径val:images/val# 验证集图像路径nc:2# 类别数names:[不规范绑扎,螺栓销钉缺失]# 类别名称转换标注格式假设标注文件是VOC格式的XML文件我们需要将它们转换为YOLO格式的TXT文件。转换脚本importxml.etree.ElementTreeasETimportosdefconvert_voc_to_yolo(voc_file,yolo_file,class_names):treeET.parse(voc_file)roottree.getroot()widthint(root.find(size/width).text)heightint(root.find(size/height).text)withopen(yolo_file,w)asf:forobjinroot.findall(object):class_nameobj.find(name).textifclass_namenotinclass_names:continueclass_idclass_names.index(class_name)bboxobj.find(bndbox)x_minfloat(bbox.find(xmin).text)y_minfloat(bbox.find(ymin).text)x_maxfloat(bbox.find(xmax).text)y_maxfloat(bbox.find(ymax).text)x_center(x_minx_max)/2.0/width y_center(y_miny_max)/2.0/height w(x_max-x_min)/width h(y_max-y_min)/height f.write(f{class_id}{x_center}{y_center}{w}{h}\n)defconvert_all_voc_to_yolo(voc_dir,yolo_dir,class_names):os.makedirs(yolo_dir,exist_okTrue)forfilenameinos.listdir(voc_dir):iffilename.endswith(.xml):voc_fileos.path.join(voc_dir,filename)yolo_fileos.path.join(yolo_dir,filename.replace(.xml,.txt))convert_voc_to_yolo(voc_file,yolo_file,class_names)if__name____main__:class_names[不规范绑扎,螺栓销钉缺失]voc_train_dirPDNDDD/labels_voc/trainyolo_train_dirPDNDDD/labels/trainconvert_all_voc_to_yolo(voc_train_dir,yolo_train_dir,class_names)voc_val_dirPDNDDD/labels_voc/valyolo_val_dirPDNDDD/labels/valconvert_all_voc_to_yolo(voc_val_dir,yolo_val_dir,class_names)YOLOv8训练代码安装YOLOv8库和依赖项gitclone https://github.com/ultralytics/ultralytics.gitcdultralytics pipinstall-rrequirements.txt训练模型fromultralyticsimportYOLOdeftrain_model(data_yaml_path,model_config,epochs,batch_size,img_size,augment):# 加载模型modelYOLO(model_config)# 训练模型resultsmodel.train(datadata_yaml_path,epochsepochs,batchbatch_size,imgszimg_size,augmentaugment)# 保存模型model.save(runs/train/power_distribution_network_defect_detection/best.pt)if__name____main__:data_yaml_pathPDNDDD/data.yamlmodel_configyolov8s.yamlepochs100batch_size16img_size640augmentTruetrain_model(data_yaml_path,model_config,epochs,batch_size,img_size,augment)详细解释安装YOLOv8和依赖项克隆YOLOv8仓库并安装所有必要的依赖项。训练模型导入YOLOv8库。加载模型配置文件。调用model.train方法进行训练。保存训练后的最佳模型。运行训练脚本将上述脚本保存为一个Python文件例如train_yolov8_pdnddd.py然后运行它。python train_yolov8_pdnddd.py评估模型评估模型fromultralyticsimportYOLOdefevaluate_model(data_yaml_path,weights_path,img_size,conf_threshold):# 加载模型modelYOLO(weights_path)# 评估模型resultsmodel.val(datadata_yaml_path,imgszimg_size,confconf_threshold)# 打印评估结果print(results)if__name____main__:data_yaml_pathPDNDDD/data.yamlweights_pathruns/train/power_distribution_network_defect_detection/best.ptimg_size640conf_threshold0.4evaluate_model(data_yaml_path,weights_path,img_size,conf_threshold)详细解释评估模型导入YOLOv8库。加载训练好的模型权重。调用model.val方法进行评估。打印评估结果。运行评估脚本将上述脚本保存为一个Python文件例如evaluate_yolov8_pdnddd.py然后运行它。python evaluate_yolov8_pdnddd.py一键运行脚本为了实现一键运行可以将训练和评估脚本合并到一个主脚本中并添加命令行参数来控制运行模式。importargparsefromultralyticsimportYOLOdefconvert_voc_to_yolo(voc_file,yolo_file,class_names):treeET.parse(voc_file)roottree.getroot()widthint(root.find(size/width).text)heightint(root.find(size/height).text)withopen(yolo_file,w)asf:forobjinroot.findall(object):class_nameobj.find(name).textifclass_namenotinclass_names:continueclass_idclass_names.index(class_name)bboxobj.find(bndbox)x_minfloat(bbox.find(xmin).text)y_minfloat(bbox.find(ymin).text)x_maxfloat(bbox.find(xmax).text)y_maxfloat(bbox.find(ymax).text)x_center(x_minx_max)/2.0/width y_center(y_miny_max)/2.0/height w(x_max-x_min)/width h(y_max-y_min)/height f.write(f{class_id}{x_center}{y_center}{w}{h}\n)defconvert_all_voc_to_yolo(voc_dir,yolo_dir,class_names):os.makedirs(yolo_dir,exist_okTrue)forfilenameinos.listdir(voc_dir):iffilename.endswith(.xml):voc_fileos.path.join(voc_dir,filename)yolo_fileos.path.join(yolo_dir,filename.replace(.xml,.txt))convert_voc_to_yolo(voc_file,yolo_file,class_names)deftrain_model(data_yaml_path,model_config,epochs,batch_size,img_size,augment):# 加载模型modelYOLO(model_config)# 训练模型resultsmodel.train(datadata_yaml_path,epochsepochs,batchbatch_size,imgszimg_size,augmentaugment)# 保存模型model.save(runs/train/power_distribution_network_defect_detection/best.pt)defevaluate_model(data_yaml_path,weights_path,img_size,conf_threshold):# 加载模型modelYOLO(weights_path)# 评估模型resultsmodel.val(datadata_yaml_path,imgszimg_size,confconf_threshold)# 打印评估结果print(results)defmain(mode):class_names[不规范绑扎,螺栓销钉缺失]data_yaml_pathPDNDDD/data.yamlmodel_configyolov8s.yamlepochs100batch_size16img_size640conf_threshold0.4augmentTrueifmodeconvert:voc_train_dirPDNDDD/labels_voc/trainyolo_train_dirPDNDDD/labels/trainconvert_all_voc_to_yolo(voc_train_dir,yolo_train_dir,class_names)voc_val_dirPDNDDD/labels_voc/valyolo_val_dirPDNDDD/labels/valconvert_all_voc_to_yolo(voc_val_dir,yolo_val_dir,class_names)elifmodetrain:train_model(data_yaml_path,model_config,epochs,batch_size,img_size,augment)elifmodeeval:weights_pathruns/train/power_distribution_network_defect_detection/best.ptevaluate_model(data_yaml_path,weights_path,img_size,conf_threshold)else:print(Invalid mode. Use convert, train, or eval.)if__name____main__:parserargparse.ArgumentParser(descriptionConvert, train, or evaluate YOLOv8 on the Power Distribution Network Defect Detection Dataset.)parser.add_argument(mode,typestr,choices[convert,train,eval],helpMode: convert, train, or eval)argsparser.parse_args()main(args.mode)详细解释命令行参数使用argparse库添加命令行参数控制脚本的运行模式转换、训练或评估。主函数根据传入的模式参数调用相应的转换、训练或评估函数。运行主脚本将上述脚本保存为一个Python文件例如main_yolov8_pdnddd.py然后运行它。转换标注格式python main_yolov8_pdnddd.py convert训练模型python main_yolov8_pdnddd.py train评估模型python main_yolov8_pdnddd.pyeval
)
