一种智慧矿山场景下的目标检测算法  

A Target Detection Algorithm in Smart Mine Scene

作　　者：姚珊珊 王静宇[1] 郝斌[1] 张飞[1] 高鹭[1] 任晓颖[1] Yao Shanshan;Wang Jingyu;Hao Bin;Zhang Fei;Gao Lu;Ren Xiaoying(School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014000,China)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014000

出　　处：《兵工自动化》2025年第4期18-25,共8页Ordnance Industry Automation

基　　金：内蒙古自治区科技计划项目(2021GG0046,2021GG0048)。

摘　　要：针对露天矿生产场景中存在着目标像素低、小目标众多、背景复杂等问题,在YOLOv5s的基础上提出一种多尺度和超分辨率网络(multiscale and super-resolution network,MS_Net)。在特征融合模块,将PANet的三尺度检测升级为四尺度检测,提高网络的多尺度学习能力,并使用子像素卷积作为上采样方法;提出一种多层融合(multi layer fusion,MLF)模块,融合了PANet 3个输出层的特征,得到一个具有丰富语义信息和空间信息的特征图;在预测层中,使用SIoU作为定位损失函数,优化模型的参数。实验结果表明:MS_Net网络在PASCALVOC数据集上mAP为79.4%,FPS为59;在矿山数据集上mAP为80.2%,FPS为64.5,模型可快速、准确、高效地对露天矿中的目标进行识别检测。Aiming at the problems of low target pixels,numerous small targets and complex background in the production scene of open-pit mine,an multiscale and super-resolution network(MS_Net)is proposed based on YOLOv5s.In the feature fusion module,the three-scale detection of PANet is upgraded to four-scale detection to improve the multi-scale learning ability of the network,and sub-pixel convolution is used as an up-sampling method;A multi layer fusion(MLF)module is proposed to fuse the features of three output layers of PANet,and a feature map with rich semantic information and spatial information is obtained.In the prediction layer,SIoU is used as the localization loss function to optimize the parameters of the model.The experimental results show that the mAP of MS_Net is 79.4%and the FPS is 59 on PASCAL VOC data set,and the mAP is 80.2%and the FPS is 64.5 on mine data set,and the model can identify and detect the target in the open-pit mine quickly,accurately and efficiently.

关 键 词：智慧矿山 目标检测 YOLOv5s 多层特征融合 子像素卷积 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]