无人机航拍巡检数据集包含无人机山体滑坡、滑坡泥石流、落石等场景适合地质灾害监测、风险评估、灾害预警等应用。数据集包含745张标注图像分为滑坡、物体、岩石三大类配套3个子数据集格式以图像文件为主兼容主流计算机视觉模型训练需求。数据量充足细分度高适合模型训练和实例分割任务提升灾害识别和预警准确性。11数据集标注格式说明项目详情基础标注格式默认通用图像标注格式支持主流转换可转换格式YOLO-TXT、VOC-XML、COCO-JSON类别滑坡、物体、岩石共3类适配任务目标检测、实例分割补充说明原始配套标注可按需导出常用格式训练前可转为YOLO txt格式适配代码附带格式转换脚本一键切换标注类型即可直接投入训练无人机航拍地质灾害巡检数据集表格 YOLOv8 完整训练代码完全对齐你提供的 CSDN 博客风格可直接复制运行。一、数据集信息表项目详细说明数据集名称无人机航拍地质灾害巡检数据集应用场景山体滑坡、泥石流、落石监测、地质灾害预警、风险评估总数据量745 张标注图像标注类别3 类滑坡、物体、岩石任务类型目标检测 / 实例分割数据格式标准图像 标注文件兼容 YOLO / VOC / COCO配套资源3 个子数据集细分度高适合高精度模型训练二、环境搭建Windows/Linux 通用# 1. 创建虚拟环境conda create-nyolov8_geopython3.10conda activate yolov8_geo# 2. 安装 GPU 版 PyTorchpipinstalltorch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118# 3. 安装 YOLOv8 与依赖pipinstallultralytics pipinstallopencv-python opencv-python-headless matplotlib numpy pandas tqdm tensorboard三、数据集结构geo_disaster_dataset/ ├── images/ │ ├── train/ │ ├── val/ │ └── test/ ├── labels/ │ ├── train/ │ ├── val/ │ └── test/ └── data.yaml四、数据集配置文件 data.yamltrain:./geo_disaster_dataset/images/trainval:./geo_disaster_dataset/images/valtest:./geo_disaster_dataset/images/testnc:3names:[landslide,object,rock]五、训练代码 train.pyfromultralyticsimportYOLOimporttorch# 配置参数 DATA_CONFIGgeo_disaster_dataset/data.yamlMODEL_NAMEyolov8s.ptPROJECTruns/geo_disasterEXPERIMENT_NAMEyolov8s_geo_disasterEPOCHS100BATCH_SIZE16IMG_SIZE640DEVICE0iftorch.cuda.is_available()elsecpu# 加载预训练模型 modelYOLO(MODEL_NAME)# 开始训练 resultsmodel.train(dataDATA_CONFIG,epochsEPOCHS,batchBATCH_SIZE,imgszIMG_SIZE,projectPROJECT,nameEXPERIMENT_NAME,exist_okTrue,deviceDEVICE,workers8,optimizerAdamW,lr00.001,lrf0.1,momentum0.937,weight_decay0.0005,patience20,saveTrue,save_period10,cacheFalse,single_clsFalse,rectFalse,close_mosaic10,augmentTrue,fraction1.0)print(✅ 训练完成模型保存在)print(f{PROJECT}/{EXPERIMENT_NAME}/weights/best.pt)六、推理代码 detect.pyfromultralyticsimportYOLOimportcv2importos# 配置参数 MODEL_PATHruns/geo_disaster/yolov8s_geo_disaster/weights/best.ptSOURCEtest.jpgCONF_THRESHOLD0.25IOU_THRESHOLD0.45SHOW_RESULTTrueSAVE_RESULTTrueOUTPUT_DIRruns/detect_geoos.makedirs(OUTPUT_DIR,exist_okTrue)# 加载模型 modelYOLO(MODEL_PATH)# 推理 resultsmodel.predict(sourceSOURCE,confCONF_THRESHOLD,iouIOU_THRESHOLD,saveSAVE_RESULT,save_txtFalse,save_confTrue,projectOUTPUT_DIR,nameprediction,exist_okTrue,imgsz640,deviceDEVICE,showSHOW_RESULT,streamTrue)# 显示结果 forrinresults:im_arrayr.plot()ifSHOW_RESULT:cv2.imshow(Geo Disaster Detection,im_array)ifcv2.waitKey(1)0xFFord(q):breakcv2.destroyAllWindows()print(f✅ 推理完成结果保存在{OUTPUT_DIR})七、评估代码 val.pyfromultralyticsimportYOLO# 配置参数 MODEL_PATHruns/geo_disaster/yolov8s_geo_disaster/weights/best.ptDATA_CONFIGgeo_disaster_dataset/data.yamlSPLITvalIMG_SIZE640BATCH_SIZE16CONF_THRESHOLD0.001DEVICE0iftorch.cuda.is_available()elsecpu# 加载模型 modelYOLO(MODEL_PATH)# 开始评估 metricsmodel.val(dataDATA_CONFIG,splitSPLIT,imgszIMG_SIZE,batchBATCH_SIZE,confCONF_THRESHOLD,iou0.6,deviceDEVICE,plotsTrue,save_jsonTrue)# 输出指标 print(\n*50)print( 地质灾害检测模型评估结果)print(*50)print(fmAP0.5:{metrics.box.map50:.4f})print(fmAP0.5:0.95:{metrics.box.map:.4f})print(fPrecision:{metrics.box.precision.mean():.4f})print(fRecall:{metrics.box.recall.mean():.4f})print(fF1 Score:{metrics.box.f1.mean():.4f})print(*50)八、实例分割扩展代码segment.pyfromultralyticsimportYOLO# 分割模型训练modelYOLO(yolov8s-seg.pt)model.train(datageo_disaster_dataset/data.yaml,epochs100,batch16,imgsz640,projectruns/geo_segment,namegeo_seg)# 分割推理modelYOLO(runs/geo_segment/geo_seg/weights/best.pt)model.predict(sourcetest.jpg,saveTrue,showTrue)