基于YOLOv7 tiny的煤岩图像检测算法  

Coal Rock Image Detection Algorithm Based on YOLOv7 Tiny

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

机构地区：[1]黑龙江科技大学计算机与信息工程学院,黑龙江哈尔滨150022

出　　处：《黑龙江工业学院学报(综合版)》2024年第12期104-107,共4页Journal of Heilongjiang University of Technology(Comprehensive Edition)

基　　金：黑龙江省省属本科高校基本科研业务费项目(项目编号: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 tiny网络结构的轻量化煤岩图像检测算法。算法引入卷积核为7的卷积模块ConvNeXt v2来替换普通卷积模块,提升煤炭特征获得效果;利用注意力机制,替换1×1大小卷积模块,改进ELAN模块,使算法提取更丰富的目标信息。结果表明:与YOLOv7tiny算法相比,改进后算法准确率提升了1.3%,召回率提升了1.0%,平均准确率提升2.7%,浮点计算量下降了1.7G,参数量降低0.93M,下降了总量的15.4%。A lightweight coal and rock image detection algorithm is proposed to improve the YOLOv7 tiny network structure by replacing some ordinary convolutional modules,in response to the difficulty in balancing accuracy and model size in current coal rock image detection and recognition.The algorithm introduces ConvNeXt v2,a convolution module with a kernel of 7,to replace the regular convolution module and improve the effectiveness of coal feature acquisition;by utilizing attention mechanism,to replacing the 1×1 convolution module and improve the ELAN module,the algorithm can extract richer target information.The results showed that compared with the YOLOv7tiny algorithm,the improved algorithm improved accuracy by 1.3%,recall rate by 1.0%,average accuracy by 2.7%,floating-point computation dropped by 1.7G,parameter quantity dropped by 0.93M,and decrease of 15.4%of the total.

关 键 词：煤岩检测 轻量化 YOLOv7tiny 注意力机制 ConvNeXt v2 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]