Cao Zhenguan, Yang Haixia, Fang Liao, Li Zhuoqin, Li Jinbiao, Dong Gaohui
School of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan 232001, Anhui, China.
Rev Sci Instrum. 2024 Sep 1;95(9). doi: 10.1063/5.0207733.
Meter reading recognition is an important link for robots to complete inspection tasks. To solve the problems of low detection accuracy and inaccurate localization of current meter reading recognition algorithms, the YOLOV7-SSWD (YOLOV7-SiLU-SimAM-Wise-IoU-DyHeads) model is proposed, a novel detection model based on the multi-head attention mechanism, which is improved on the YOLOV7-Tiny model. First, the Wise-IoU loss function is used to solve the problem of sample quality imbalance and improve the model's detection accuracy. Second, a new convolutional block is constructed using the SiLU activation function and applied to the YOLOV7-Tiny model to enhance the model's generalization ability. The dynamic detection header is then built as the header of YOLOV7-Tiny, which realizes the fusion of multi-scale feature information and improves the target recognition performance. Finally, we introduce SimAM to improve the feature extraction capability of the network. In this paper, the importance of each component is fully verified by ablation experiments and comparative analysis. The experiments showed that the mAP and F1-scores of the YOLOV7-SSWD model reached 89.8% and 0.84. Compared with the original network, the mAP increased by 8.1% and the F1-scores increased by 0.1. The YOLOV7-SSWD algorithm has better localization and recognition accuracy and provides a reference for deploying inspection robots to perform automatic inspections.
抄表识别是机器人完成巡检任务的重要环节。针对当前抄表识别算法检测精度低、定位不准确的问题，提出了YOLOV7-SSWD（YOLOV7-SiLU-SimAM-Wise-IoU-DyHeads）模型，这是一种基于多头注意力机制的新型检测模型，在YOLOV7-Tiny模型基础上进行了改进。首先，使用Wise-IoU损失函数解决样本质量不平衡问题，提高模型检测精度。其次，利用SiLU激活函数构建新的卷积块并应用于YOLOV7-Tiny模型，增强模型泛化能力。然后构建动态检测头作为YOLOV7-Tiny的头部，实现多尺度特征信息融合，提高目标识别性能。最后，引入SimAM提高网络特征提取能力。本文通过消融实验和对比分析充分验证了各组件的重要性。实验表明，YOLOV7-SSWD模型的mAP和F1分数分别达到89.8%和0.84。与原网络相比，mAP提高了8.1%，F1分数提高了0.1。YOLOV7-SSWD算法具有更好的定位和识别精度，为部署巡检机器人进行自动巡检提供了参考。
