改进YOLOE算法在压板状态检测中的应用

The Application of Improved YOLOE Algorithm for Platen State Detection

摘要: 针对当前变电站压板种类多、数量大、小目标易误检，及压板状态检测的速度慢和准确性不高等问题，本文设计实现了一种基于深度学习的压板状态检测系统。以YOLOE检测算法为基础，在其主干网络中融入ECA(Efficient Channel Attention)注意力机制，强化网络的特征提取能力，以提升模型检测的准确度。引入SIoU代替原损失函数，以解决检测精度低和小目标易漏检误检的问题，并且提高模型的训练速度。对比实验结果表明改进后的YOLOE算法在变电站压板检测中上取得了90.3%的平均检测精度、96.9%的召回率以及97.8%的精确率，能够满足变电站运行检测要求。

Abstract: Aiming at the current problems of multiple types and large number of substation voltage plates, easy misdetection of small targets, and slow speed and low accuracy of platen state detection, this paper designs and implements a deep learning-based platen state detection system. Based on the YOLOE detection algorithm, the ECA (Efficient Channel Attention) attention mechanism is integrated into its backbone network to strengthen the feature extraction capability of the network in order to improve the accuracy of model detection. SIoU is introduced to replace the original loss function to solve the problems of low detection accuracy and easy to miss and error detection of small targets and to improve the training speed of the model. Comparative experimental results show that the improved YOLOE algorithm achieves an average detection precision of 90.3%, a recall rate of 96.9% and an accuracy rate of 97.8% in substation voltage plate detection, which can meet the requirements of substation operation detection.