Chongqing University of Posts and Telecommunications
针对复杂矿井环境下光照度低、目标尺度变化大、目标间遮挡严重，现有的目标检测网络特征提取困难、检测效果差等问题，提出了改进的S3-YOLOv5s的矿井人员防护设备检测算法。在主干网络中加入无参注意力模块SimAM(A Simple, Parameter-Free Attention Module)，提升网络的特征提取能力；引入尺度均衡特征金字塔卷积(Scale-Equalizing Pyramid Convolution，SEPC)，加强多尺度特征融合；最后采用SIoU作为边框回归损失函数并使用K-means++算法进行先验锚框聚类，提高边框检测精度。实验表明，相比现有的YOLOv5s算法，本文算法在所有类别的平均检测精确度从89.64%提升到了92.86%，在复杂矿井环境条件下对人员防护设备有优良的检测能力，验证了所提方法的有效性。
Aiming at the problems of low illumination, large change of target scale, serious occlusion between targets, difficult feature extraction of existing target detection network, poor detection effect, etc. in complex mine environment, an improved S3-YOLOv5s mine personnel protection equipment detection algorithm is proposed. A simple, parameter free attention module (SimAM) is added to the backbone network to improve the network""s feature extraction capability; Scale Equalizing Pyramid Convolution (SEPC) is introduced to strengthen multi-scale feature fusion; Finally, SIoU is used as the frame regression loss function and K-means++algorithm is used for prior anchor frame clustering to improve the frame detection accuracy. The experiment shows that, compared with the YOLOv5s algorithm, the average detection accuracy of this algorithm in all categories is improved from 89.64% to 92.86%, and the algorithm has excellent detection capability for personnel protection equipment under complex mine environment conditions, which verifies the effectiveness of the proposed method.