边缘设备上的葡萄园田间场景障碍检测  

Vineyard scene obstacle detection based on edge devices

作　　者：崔学智 冯全 王书志[2] 张建华[3] Cui Xuezhi;Feng Quan;Wang Shuzhi;Zhang Jianhua(School of Mechanical and Electrical Engineering,Gansu Agriculture University,Lanzhou,730070,China;College of Electrical Engineering,Northwest University for Nationalities,Lanzhou,730030,China;Agricultural Information Institute of CAAS,Beijing,100081,China)

机构地区：[1]甘肃农业大学机电工程学院,兰州市730070 [2]西北民族大学机电工程学院,兰州市730030 [3]中国农业科学院农业信息研究所,北京市100081

出　　处：《中国农机化学报》2021年第9期150-156,共7页Journal of Chinese Agricultural Mechanization

基　　金：中央高校基本科研业务费项目(31920200043);国家自然基金面上项目(31971792)。

摘　　要：为在无人驾驶农机上实现快速、准确的葡萄园田间障碍目标检测,将EfficientDet D0、YOLOV4 TINY、YOLOV3 TINY、YOLO-FASTEST四种轻量级目标检测神经网络在自建的葡萄园田间场景数据集进行训练,将训练模型移植到边缘设备NVIDIA JETSON TX2(以下简称TX2)上,对这4种模型的障碍检测精度和在TX2上的适用性进行试验评估。试验结果表明,YOLOV3 TINY、YOLOV4 TINY、EfficientDet D0、YOLO-FASTEST对葡萄园田间场景障碍检测平均精度mAP分别为0.648、0.601、0.598和0.401。在TX2的实测结果显示,上述网络模型实时视频检测帧率分别为34.24帧、24.75帧、2.34帧和2.97帧。4种目标检测网络中,YOLOV3 TINY在数据集上的检测精度最高、实时检测速度最快,但对硬件资源消耗也相对较高。而在考虑硬件资源消耗时,使用YOLOV4 TINY可以在检测精度、实际运行速度和硬件资源消耗之间维持更好的平衡性,同时可以在运行多任务的情况下取得好的效果。In order to achieve fast and accurate obstacle detection in vineyard field on unmanned agricultural machinery,four kinds of lightweight target detection neural networks,EfficientDet D0,YOLOV4 TINY,YOLOV3 TINY,and YOLO-FASTEST,were trained with the independently built vineyard field scene dataset,and the training model eretransplanted to the edge device NVIDIA JETSON TX2.The accuracy of obstacle detection and the applicability of the four models on TX2 were evaluated.The results showed that the mAP of YOLOV3 TINY,YOLOV4 TINY,EfficientDet D0,and YOLO-FASTEST were 0.648,0.601,0.598,and 0.401,respectively.The experimental results on TX2 showed that the real-time video detection frame rates of the above network models were 34.24 frames,24.75 frames,2.34 frames,and 2.97 frames,respectively.Among the four target detection networks,YOLOV3 TINY has the highest detection accuracy on the dataset and the fastest real-time detection speed,but it also consumes high hardware resources relatively.When considering the hardware resource consumption,YOLOV4 TINY can better balance detection accuracy,running speed,and hardware resource consumption and achieve good results when running multiple tasks.

关 键 词：田间场景 轻量级网络模型 边缘设备 实时检测 适用性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]