摘要
传统的图像处理方法难以有效应对复杂背景和不同尺度的缺陷,文章提出了一种融合双向特征金字塔网络(BiFPN)的YOLOv8架构优化策略,旨在提升皮革缺陷识别的精度和效率。YOLOv8作为一种高效的目标检测框架,结合BiFPN的多尺度特征融合优势,增强了模型在复杂背景下的特征提取能力。通过在YOLOv8中引入BiFPN模块,模型能够更好地捕捉不同尺度的皮革缺陷,并通过优化后的损失函数进一步提高识别的准确性和稳定性。实验结果表明,改进前的YOLOv8权重为6.3 MB,改进后降至4.3 MB,且mAP50提高了0.2%。该优化策略相较于传统方法和未融合BiFPN的YOLOv8,提升了识别精度和识别速度,优化了YOLOv8框架在皮革缺陷检测中的有效性及实际应用潜力。
Traditional image processing methods are difficult to effectively deal with complex backgrounds and defects with different scales.This paper proposed a YOLOv8 architecture optimization strategy that integrates Bidirectional Feature Pyramid Network(BiFPN)to improve the accuracy and efficiency of leather defect recognition.YOLOv8,as an efficient object detection framework,combined the multi-scale feature fusion advantages of BiFPN to enhance the model's feature extraction ability in complex backgrounds.By introducing the BiFPN module in YOLOv8,the model can better capture leather defects with different scales,and further improve the accuracy and stability of recognition through the optimized loss functions.The experimental results showed that the weight of YOLOv8 before improvement was 6.3 MB,which decreased to 4.3 MB after improvement,and mAP50 increased by 0.2%.This optimization strategy improves recognition accuracy and speed compared to traditional methods and YOLOv8 without BiFPN fusion,and optimizes the effectiveness and practical application potential of the YOLOv8 framework in leather defect detection.
作者
唐灏
陈法明
冯志鹏
何凌志
TANG Hao;CHEN Faming;FENG Zhipeng;HE Lingzhi(Jiangmen Polytechnic,Jiangmen 529030,China;College of Information Science and Technology,Jinan University,Guangzhou 510632,China;Guangzhou Qingbei Intelligent Technology Co.,Ltd,Guangzhou 510450,China;Jiangsu Juyi Robot Co.,Ltd,Nantong 211599,China;Jiangsu Normal University,Guangzhou 510450,China)
出处
《皮革科学与工程》
北大核心
2025年第5期22-30,60,共10页
Leather Science and Engineering
