摘 要:针对目标检测算法 YOLOv3 检测精度低、目标识别效果差等问题,从特征提取和特征融合的角度提出一种改进的 YOLOv3 目标检测算法。采取连续残差结构和深度卷积双路特征提取来扩展感受野,在深度卷积模块中以改进的混合池化来替换最大池化;在特征融合方面,引入 CBAM,并在增强残差模块中增加了注意力特征融合模块。实验结果表明,改良后的YOLOv3 算法在百度与北京林业大学合作的 Insects 昆虫数据集上的检测精度达到了 71.22%,比原始算法的检测精度提升 4.88个百分点,验证了该算法的有效性。
关键词:目标检测;YOLOv3;注意力机制;昆虫;CBAM
DOI:10.19850/j.cnki.2096-4706.2022.16.019
基金项目:安徽省教育厅科研项目(KJ2017A085);安徽省自然科学基金项目(1808085QF205)
中图分类号:TP391.4 文献标识码:A 文章编号:2096-4706(2022)16-0071-04
Research on Target Detection Method Based on Improved YOLOv3
WANG Jiqian, LIU Huanhuan, LIAO Tao, ZHU Xiaodong
(Anhui University of Science and Technology, Huainan 232001, China)
Abstract: Aiming at the problems of low detection accuracy and poor target recognition effect of target detection algorithm YOLOv3, an improved YOLOv3 target detection algorithm is proposed from the perspective of feature extraction and feature fusion. The continuous residual structure and deep convolution two-way feature extraction are adopted to expand the receptive field, and the improved mixed pooling is used to replace the maximum pooling in the deep convolution module; in the aspect of feature fusion, CBAM is introduced, and the attention feature fusion module is added to the enhanced residual module. The experimental results show that the detection accuracy of the improved YOLOv3 algorithm on the Insects insect data set cooperated by Baidu and Beijing Forestry University reaches 71.22%, which is 4.88% higher than the detection accuracy of the original algorithm, which verifies the effectiveness of the algorithm.
Keywords: target detection; YOLOv3; attention mechanism; insect; CBAM
参考文献:
[1] 邓敏,黄世醒,黄燕娟,等 . 基于 YOLOV3 模型的甘蔗丛环境下行人检测方法 [J]. 农机化研究,2022,45(1):8-14+57.
[2] 马书浩,安居白 . 基于 YOLOv3 改进的肺炎检测算法 [J].激光与光电子学进展,2020,57(18):318-324.
[3] 史梦安,蔡慧敏,陆振宇 . 基于 YOLOv3 改进的轻量化人脸检测方法 [J]. 计算机工程与设计,2022,43(3):858-865.
[4] 刘汉生 . 陷阱式储粮害虫信息采集终端及其系统的研究与实现 [D]. 北京:北京邮电大学,2018.
[5] MOUSAVI S F,ABBASPOUR-FARD M H,AGHKHANI M H, et al. Acoustic detection possibility of different stages of the confused flourbeetle (Triboiumconfusum) in grain bulks using an audio sensor[J]. Journal of Agricultural Science and Technology,2017,19:1551-1563.
[6] 牛浩青,欧鸥,饶姗姗,等 . 改进 YOLOv3 的遥感影像小目标检测方法 [J]. 计算机工程与应用,2022,58(13):241-248.
[7] 杨文姬,李浩,王映龙,等 . 改进 YOLOv3 的多尺度高分辨率特征增强图像目标检测 [J/OL]. 小型微型计算机系统:1-8[2022-04-22].http://kns.cnki.net/kcms/detail/21.1106.tp.20220418.1307.014.html.
[8] 张丽莹,庞春江,王新颖,等 . 基于改进 YOLOv3 的多尺度目标检测算法 [J/OL]. 计算机应用:1-11[2022-04-03].http:// kns.cnki.net/kcms/detail/51.1307.TP.20211223.1844.016.html.
作者简介:王继千(1996—),男,汉族,安徽宿州人,硕士研究生在读,研究方向:目标检测;刘唤唤(1984—),女,汉族,安徽淮南人,讲师、博士,研究方向:金属纳米材料、多孔硅材料、纳米生化传感器的设计与测试,数据挖掘;廖涛(1977—),男,汉族,安徽淮南人,副教授、博士,研究方向:Web 数据挖掘、智能信息处理;朱小东(1980—),男,汉族,安徽淮南人,讲师、博士,研究方向:模式识别、并行算法。