智慧工业巡检-基于深度学习YOLO26算法的矿山煤炭输送场景目标检测传送带物料异常检测1、矿山煤炭输送异常检测系统YOLOv26 版完整实现支持实时视频流检测、多类异物识别、报警输出可直接部署在工业监控系统中。一、系统功能说明模块功能目标检测识别传送带煤炭中的金属异物螺栓、钢筋、铁块、螺母等实时监控接入工业摄像头/视频流毫秒级检测报警机制检测到异物后可触发声光报警、保存截图/视频片段日志记录自动记录异常时间、位置、异物类型部署方式支持本地PC、边缘设备、ROS工业系统集成二、数据集与模型配置1. 数据集结构coal_conveyor_dataset/ ├── images/ │ ├── train/ │ ├── val/ │ └── test/ └── labels/ ├── train/ ├── val/ └── test/2. 类别定义煤炭异物类别ID英文标签中文标签0bolt螺栓/螺母1rebar钢筋/螺纹钢2metal_sheet金属片/铁板3other_metal其他金属异物3. YOLOv26 配置文件coal_anomaly.yamlpath:./coal_conveyor_datasettrain:images/trainval:images/valtest:images/testnc:4names:0:bolt1:rebar2:metal_sheet3:other_metal三、YOLOv26 训练代码fromultralyticsimportYOLOif__name____main__:# 加载YOLOv26模型推荐使用n/s版工业场景实时性优先modelYOLO(yolo26n.pt)# 工业场景训练参数优化resultsmodel.train(datacoal_anomaly.yaml,epochs150,imgsz640,batch16,device0,lr00.01,lrf0.01,warmup_epochs3,cos_lrTrue,patience15,cacheTrue,ampTrue,augmentTrue,# 针对传送带场景的数据增强hsv_h0.01,hsv_s0.5,hsv_v0.3,degrees10,perspective0.001,fliplr0.5,mosaic1.0,mixup0.1,namecoal_conveyor_anomaly)# 测试集评估model.val(splittest)四、传送带实时检测主程序带报警功能importcv2importtimefromdatetimeimportdatetimefromultralyticsimportYOLOclassCoalConveyorDetector:def__init__(self,model_pathbest.pt,conf_threshold0.3):self.modelYOLO(model_path)self.conf_thresholdconf_threshold self.alert_cooldown5# 报警冷却时间秒self.last_alert_time0deftrigger_alert(self,frame,label,confidence):触发报警保存截图、打印日志可扩展为声光/继电器输出current_timetime.time()ifcurrent_time-self.last_alert_timeself.alert_cooldown:returnself.last_alert_timecurrent_time timestampdatetime.now().strftime(%Y%m%d_%H%M%S)cv2.imwrite(falerts/alert_{timestamp}.jpg,frame)print(f⚠️ 异常检测{label}置信度{confidence:.2f}时间{datetime.now()})# 此处可扩展调用串口控制继电器、触发声光报警、发送报警信息defprocess_frame(self,frame):处理单帧图像返回标注后的帧和检测结果resultsself.model(frame,confself.conf_threshold)annotated_frameresults[0].plot()detected_objects[]forboxinresults[0].boxes:clsint(box.cls[0])labelresults[0].names[cls]confidencefloat(box.conf[0])detected_objects.append((label,confidence))self.trigger_alert(frame,label,confidence)returnannotated_frame,detected_objectsdefrun(self,source0,save_videoFalse): 运行实时检测 :param source: 摄像头ID/视频路径 :param save_video: 是否保存检测视频 capcv2.VideoCapture(source)fpscap.get(cv2.CAP_PROP_FPS)widthint(cap.get(cv2.CAP_PROP_FRAME_WIDTH))heightint(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))ifsave_video:fourcccv2.VideoWriter_fourcc(*mp4v)outcv2.VideoWriter(coal_anomaly_output.mp4,fourcc,fps,(width,height))whilecap.isOpened():ret,framecap.read()ifnotret:breakannotated_frame,objectsself.process_frame(frame)cv2.imshow(Coal Conveyor Anomaly Detection,annotated_frame)ifsave_video:out.write(annotated_frame)ifcv2.waitKey(1)0xFFord(q):breakcap.release()ifsave_video:out.release()cv2.destroyAllWindows()if__name____main__:detectorCoalConveyorDetector(model_pathruns/detect/coal_conveyor_anomaly/weights/best.pt)detector.run(sourceconveyor_video.mp4,save_videoTrue)五、系统部署与扩展1. 安装依赖pipinstallultralytics opencv-python numpy2. 运行方式摄像头实时检测将source0传入detector.run()视频文件检测传入视频路径如sourceconveyor_video.mp4工业相机接入替换为工业相机的IP地址或SDK调用3. 工业级扩展功能报警输出通过pyserial控制串口继电器触发现场声光报警边缘部署模型转换为ONNX/TensorRT格式部署到NVIDIA Jetson系列设备日志上传将报警数据推送到工业SCADA系统或云端平台多摄像头支持使用多线程同时处理多路视频流
基于深度学习YOLO26算法的矿山煤炭输送场景目标检测传送带物料异常检测
智慧工业巡检-基于深度学习YOLO26算法的矿山煤炭输送场景目标检测传送带物料异常检测1、矿山煤炭输送异常检测系统YOLOv26 版完整实现支持实时视频流检测、多类异物识别、报警输出可直接部署在工业监控系统中。一、系统功能说明模块功能目标检测识别传送带煤炭中的金属异物螺栓、钢筋、铁块、螺母等实时监控接入工业摄像头/视频流毫秒级检测报警机制检测到异物后可触发声光报警、保存截图/视频片段日志记录自动记录异常时间、位置、异物类型部署方式支持本地PC、边缘设备、ROS工业系统集成二、数据集与模型配置1. 数据集结构coal_conveyor_dataset/ ├── images/ │ ├── train/ │ ├── val/ │ └── test/ └── labels/ ├── train/ ├── val/ └── test/2. 类别定义煤炭异物类别ID英文标签中文标签0bolt螺栓/螺母1rebar钢筋/螺纹钢2metal_sheet金属片/铁板3other_metal其他金属异物3. YOLOv26 配置文件coal_anomaly.yamlpath:./coal_conveyor_datasettrain:images/trainval:images/valtest:images/testnc:4names:0:bolt1:rebar2:metal_sheet3:other_metal三、YOLOv26 训练代码fromultralyticsimportYOLOif__name____main__:# 加载YOLOv26模型推荐使用n/s版工业场景实时性优先modelYOLO(yolo26n.pt)# 工业场景训练参数优化resultsmodel.train(datacoal_anomaly.yaml,epochs150,imgsz640,batch16,device0,lr00.01,lrf0.01,warmup_epochs3,cos_lrTrue,patience15,cacheTrue,ampTrue,augmentTrue,# 针对传送带场景的数据增强hsv_h0.01,hsv_s0.5,hsv_v0.3,degrees10,perspective0.001,fliplr0.5,mosaic1.0,mixup0.1,namecoal_conveyor_anomaly)# 测试集评估model.val(splittest)四、传送带实时检测主程序带报警功能importcv2importtimefromdatetimeimportdatetimefromultralyticsimportYOLOclassCoalConveyorDetector:def__init__(self,model_pathbest.pt,conf_threshold0.3):self.modelYOLO(model_path)self.conf_thresholdconf_threshold self.alert_cooldown5# 报警冷却时间秒self.last_alert_time0deftrigger_alert(self,frame,label,confidence):触发报警保存截图、打印日志可扩展为声光/继电器输出current_timetime.time()ifcurrent_time-self.last_alert_timeself.alert_cooldown:returnself.last_alert_timecurrent_time timestampdatetime.now().strftime(%Y%m%d_%H%M%S)cv2.imwrite(falerts/alert_{timestamp}.jpg,frame)print(f⚠️ 异常检测{label}置信度{confidence:.2f}时间{datetime.now()})# 此处可扩展调用串口控制继电器、触发声光报警、发送报警信息defprocess_frame(self,frame):处理单帧图像返回标注后的帧和检测结果resultsself.model(frame,confself.conf_threshold)annotated_frameresults[0].plot()detected_objects[]forboxinresults[0].boxes:clsint(box.cls[0])labelresults[0].names[cls]confidencefloat(box.conf[0])detected_objects.append((label,confidence))self.trigger_alert(frame,label,confidence)returnannotated_frame,detected_objectsdefrun(self,source0,save_videoFalse): 运行实时检测 :param source: 摄像头ID/视频路径 :param save_video: 是否保存检测视频 capcv2.VideoCapture(source)fpscap.get(cv2.CAP_PROP_FPS)widthint(cap.get(cv2.CAP_PROP_FRAME_WIDTH))heightint(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))ifsave_video:fourcccv2.VideoWriter_fourcc(*mp4v)outcv2.VideoWriter(coal_anomaly_output.mp4,fourcc,fps,(width,height))whilecap.isOpened():ret,framecap.read()ifnotret:breakannotated_frame,objectsself.process_frame(frame)cv2.imshow(Coal Conveyor Anomaly Detection,annotated_frame)ifsave_video:out.write(annotated_frame)ifcv2.waitKey(1)0xFFord(q):breakcap.release()ifsave_video:out.release()cv2.destroyAllWindows()if__name____main__:detectorCoalConveyorDetector(model_pathruns/detect/coal_conveyor_anomaly/weights/best.pt)detector.run(sourceconveyor_video.mp4,save_videoTrue)五、系统部署与扩展1. 安装依赖pipinstallultralytics opencv-python numpy2. 运行方式摄像头实时检测将source0传入detector.run()视频文件检测传入视频路径如sourceconveyor_video.mp4工业相机接入替换为工业相机的IP地址或SDK调用3. 工业级扩展功能报警输出通过pyserial控制串口继电器触发现场声光报警边缘部署模型转换为ONNX/TensorRT格式部署到NVIDIA Jetson系列设备日志上传将报警数据推送到工业SCADA系统或云端平台多摄像头支持使用多线程同时处理多路视频流
