基于RT-DETR改进的皮带运输机异物识别方法

Belt Conveyor Foreign Object Recognition Method Based on Improved RT-DETR

凭借兼顾检测精度与速度的特点,YOLO近年来已成为煤炭等工业领域目标检测模型的佼佼者。然而,YOLO的检测性能受到置信度阈值和非极大值抑制阈值等超参数设定的影响。因此,本研究提出了一种基于改进的RT-D ETR带式输送机非煤异物检测模型。该模型无需置信度过滤和非极大值抑制,从而提升了检测精度。此外,针对RT-DETR参数量较大、难以在计算资源有限的边缘设备上部署的问题,我们设计了一种EMA-Faster Net骨干网络,并将颈部网络的AIFI模块替换为LPE-AIFI模块。最后,我们采用TensorRT进行加速,并将模型部署到Jetson Orin Nano边缘计算设备上。实验结果表明,改进后的RT-DETR模型与具有相似参数量的YOLOv8s相比,其召回率高出5.6%,平均类别精度高出4.3%;经TensorRT加速后,模型帧率可达26.4 FPS,满足了实时监测的要求。

As an excellent object detection model,YOLO has recently been widely recognized in the fields of industry,such as the coal industry,due to its balance between detection accuracy and speed.However,the detection performance of YOLO is affected by the hyper-parameter settings such as confidence threshold and non-maximum suppression threshold.To solve this problem,we propose a belt conveyor non-coal foreign object detection model based on improved RT-DETR.By eliminating the process of confidence filtering and non-maximum suppression,the detection accuracy is improved.Furthermore,aiming at the problem that RT-DETR has a large number of parameters,which makes it difficult to deploy on edge devices with limited computing resources,we designed an EMA-Faster Net backbone network and replaced the AIFI module in the neck network with the LPE-AIFI module.Finally,we use TensorRT for acceleration and deploy the model on the Jetson Orin Nano edge computing device.The experimental results show that the improved RT-DETR model has 5.6%higher recall rate and 4.3%higher mean average precision compared with the YOLOv8s with similar number of parameters;after TensorRT acceleration,the model frame rate can reach 26.4 FPS,meeting the requirement of real-time monitoring.