基于改进的YOLOv8布匹残损检测算法  

Fabric Detect Detection with Improved YOLOv8

作　　者：王千 王国栋 李军 WANG Qian;WANG Guodong;LI Jun(College of Computer Science&Technology,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2024年第2期47-52,共6页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：青岛市自然科学基金资助项目(23-2-1-163-zyyd-jch)。

摘　　要：针对布匹残损检测实时性差、残损与背景相似度高及尺寸差异大等问题,提出一种基于YOLOv8的布匹残损检测算法。采用GSConv轻量化骨干网络,在保持准确性的同时减轻计算负担;引入双向特征金字塔网络,双向跨尺度连接和加权多尺度特征融合;TransViBlock模块使模型具有全局上下文感知能力,显著提升了不同尺度目标的检测效果。相较于基准模型YOLOv8,本算法准确率、召回率、平均精度分别提高2.7%、2.4%、2.9%,与传统布匹检测算法相比,在残损种类复杂和检出难度较大等情况准确性较高。A fabric damage detection algorithm based on YOLOv8 was proposed to address issues such as poor real-time performance,high similarity between damage and background,and large differences in size.The GSConv lightweight backbone network was employed to alleviate computational burden while maintaining accuracy.The Bidirectional Feature Pyramid Network was introduced to achieve efficient bidirectional cross-scale connections and weighted multi-scale feature fusion.The TransViBlock module endows the model with global contextual awareness,significantly enhancing the detection performance of targets at different scales.Compared to the baseline model YOLOv8,the accuracy,recall rate,and average precision are improved by 2.7%,2.4%,and 2.9%,respectively.Compared with traditional fabric detection algorithms,this algorithm exhibits higher accuracy in situations with complex damage types and high detection difficulty.

关 键 词：布匹残损检测 特征金字塔 全局上下文感知 轻量化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]