改进YOLOv8n的小目标检测算法
Small object detection algorithm based on improved YOLOv8n
赵金宪 1赵志滢1
作者信息
- 1. 黑龙江科技大学 计算机与信息工程学院,哈尔滨 150022
- 折叠
摘要
针对复杂场景中目标特征不明显易出现漏检和误检的问题,提出了YOLOv8n_Y小目标检测算法.通过YOLOv8n模型的骨干网络引入可变形卷积模块,在目标区域感受野有限的情况下,可变形卷积模块可以自适应地调整卷积核大小,提取更多的特征信息,将 Neck 模块添加了CBAM注意力机制,注意力机制会集中在更为重要的物体特征上,减少了误检的概率,提升对小目标检测的准确率.结果表明,YOLOv8n_Y模型在小目标吸烟数据集上的精度提升了3.3%.
Abstract
This paper proposes a YOLOv8n_Y small object detection algorithm as a solusion to the is-sues of missed and false detections due to the inapparent target features in complex real scene.The study is accomplished by incorporating deformable convolution modules into the backbone network of the YOLOv8n model;adaptively adjusting the convolution kernel size by the deformable convolution module in situations with limited receptive fields in the target area to extract more feature information;adding CBAM attention mechanism to the Neck module,which has the attention mechanism focused on more im-portant object featur,as reducing the probability of false detections and improving the accuracy of small object detection.The results indicate that YOLOv8n_Y model achieves an improvement of the accuracy by 3.3%in the small target smoking dataset.
关键词
YOLOv8n/可变形卷积模块/注意力机制/小目标检测Key words
YOLOv8n/deformable convolution module/attention mechanism/small object detec-tion引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据