改进YOLOv7的煤岩图像检测算法  

Coal rock image detection algorithm based on improved YOLOv7

作　　者：赵艳芹 邓虎诚 Zhao Yanqin;Deng Hucheng(School of Computer&Information Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学计算机与信息工程学院,哈尔滨150022

出　　处：《黑龙江科技大学学报》2024年第1期157-162,共6页Journal of Heilongjiang University of Science And Technology

基　　金：黑龙江省省属高等学校基本科研业务费项目(2022-KYYWF-0565)。

摘　　要：针对现阶段煤岩图像检测识别中精度和模型规模难以平衡的问题,提出了一种通过替换部分普通卷积模块来改进YOLOv7网络结构的煤岩图像检测算法。通过引入卷积核为7的卷积模块ConvNeXt来替换普通的3×3大小卷积模块,提升煤炭特征获得效果。利用SimAM注意力机制,替换1×1大小卷积模块,给出MP_SAM模块,使算法提取更丰富的目标信息,运用αIoU优化损失函数,使之更适用于清晰度不够高的煤岩图像,增强算法的泛化能力。结果表明,与YOLOv7算法相比,该算法的准确率提升了3.9%,mAP提升了1.5%,模型整体FLOPs减少了0.7 G,通过更小的模型,获得了更好的检测结果。This paper proposes a coal rock image detection algorithm for improving YOLOv7 network structure by replacing some ordinary convolution modules,which is designed to address the problem that is hard to balance the accuracy and model scale in current coal rock image detection.The study is accomplished by introducing ConvNeXt with a convolution kernel by 7 to replace the ordinary convolution module with the size 3×3 for improving the coal characteristics and obtaining the effect;using SimAM attention mechanism to replace convolutional modules with the size 1×1 for creating MP_SAM modules to enable the algorithm to extract more target information;optimizing the loss function by usingαIoU to make it more suitable for coal rock images with insufficient clarity,and enhance the generalization ability of the algorithm.The experimental results show that,compared with YOLOv7 algorithm,the accuracy of the algorithm increases by 3.9%,the mAP increases by 1.5%,the overall FLOPs of the model are reduced by 0.7 G,and the detection results are better obtained by using the smaller model.

关 键 词：煤岩检测 YOLOv7 SimAM ConvNeXt αIoU 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TD67[自动化与计算机技术—计算机科学与技术]