基于改进YOLOv8模型的巡检机器人目标检测方法研究

Improved YOLOv8 model-based object detection method for inspection robot

目的:针对巡检机器人对指针式仪表识别准确率不够高、对被遮挡仪表识别效果较差的问题,提出一种基于改进YOLOv8模型的巡检机器人目标检测方法。方法:首先,以YOLOv8n模型作为基础目标检测模型,在此基础上引入坐标注意力机制,加强模型对输入数据的空间结构理解;其次,将损失函数由完整交并比(complete IoU,CIoU)损失函数替换为高效交并比(efficient IoU,EIoU)损失函数,加快模型检测框的收敛速度;最后,采用柔性非极大值抑制(soft non-maximum suppression,Soft-NMS)函数替代传统的非极大值抑制(non-maximum suppression,NMS)函数,以更加平滑地抑制冗余的边界框,进一步提高检测的准确率。为验证改进模型对检测目标的识别效果,将YOLOv8n模型与改进后的YOLOv8模型(YOLOv8nxt模型)进行对比。结果:与YOLOv8n模型相比,YOLOv8nxt模型的位置损失值降低了1.3%,mAP_0.5:0.95提高了1.7%,检测准确率提高了0.87%,模型大小仅为6.2 M,检测时间仅增加了0.2 ms。结论:基于改进YOLOv8模型的巡检机器人目标检测方法提升了巡检机器人在运动过程中对仪表的识别精度和速度,能有效解决巡检机器人在目标检测阶段存在的问题。

Objective To propose a object detection method based on an improved YOLOv8 model to solve the problems of the inspection robot in low accuracy for recognizing pointer-type or obscured meters.Methods Firstly,a YOLOv8 model was chosen as the foundation object detection model,based on which the coordinate attention(CA)mechanism was introduced to enhance the model's understanding of the spatial structure of the input data over long distances;secondly,the original complete IoU(CIoU)loss function was replaced by an efficient IoU(EIoU)loss function to accelerate the convergence of the model's detection frame;finally,the soft non-maximum suppression(Soft-NMS)function took the place of the traditional NMS method to suppress the redundant bounding box smoothly and further improve the detection accuracy.The improved YOLOv8 model(YOLOv8nxt model)was compared with the YOLOv8n model to verify its efficacy for object detection.Results The YOLOv8nxt model with a size of 6.2 M had the position loss decreased by 1.3%,mAP_0.5:0.95 increased by 1.7%,detection accuracy raised by 0.87%and detection time prolonged by only 0.2 ms when comparted with the YOLOv8n model.Conclusion The improved YOLOv8 model-based object detection method enhances the accuracy and speed of the inspection robot's recognition of meters during movement,and can effectively solve the problems of the inspection robot in the object detection stage.