智能系统学报2026，Vol.21Issue(2)：353-364,12.DOI:10.11992/tis.202503010
面向失效增强和改进YOLOv8的目标检测
A failure enhancement and improvement of YOLOv8 for target detection
摘要
Abstract
To address the issues of low detection performance and weak generalization ability in target detection under complex background conditions such as illumination,weather,and occlusion,this paper proposes an improved object detection algorithm based on failure augmentation and enhanced YOLOv8(AS_YOLO).First,a variety of target unit datasets were constructed based on complex military scenarios,and an image failure augmentation technique tailored to the application environment was developed.Second,a channel-spatial parallel attention mechanism was introduced to simultaneously focus on feature and position information of targets in complex environments.Then,the AFPN structure was used to enhance feature fusion of non-adjacent hierarchical layers.Finally,the Inner_IoU loss function was adopted to address the generalization limitations of existing IoU loss functions in different detection tasks.Transfer experiments were conducted on the WSODD multi-target dataset.The experimental results show that the improved algorithm achieves an mAP0.5 of 94.0％,a 12.5 percentage point improvement over the baseline YOLOv8n model,and an mAP0.95 of 72.5％,a 15.7 percentage point improvement,indicating superior detection performance.关键词
计算机视觉/复杂环境/目标检测/YOLO/图像增强/注意力机制/特征融合/损失函数Key words
computer vision/complex environment/object detection/YOLO/image enhancement/attention mechanism/feature fusion/loss function分类
信息技术与安全科学引用本文复制引用
储文娟,李震,黄炜嘉,王宇轩..面向失效增强和改进YOLOv8的目标检测[J].智能系统学报,2026,21(2):353-364,12.基金项目
国家自然科学基金项目(62276285) （62276285）
教育部学位与研究生教育发展中心主题案例库项目(ZT-231028914) （ZT-231028914）
江苏省研究生科研与实践创新计划项目(KYCX24-4178) （KYCX24-4178）
中国科学院软件研究所合作项目(220507-2325). （220507-2325）
