基于改进的EfficientDet的布匹疵点识别  

Fabric defect recognition based on improved EfficientDet network

作　　者：杨连贺[1] 张超[1] YANG Lianhe;ZHANG Chao(School of Computer Science and Technology,Tiangong University,Tianjin 300387,China)

机构地区：[1]天津工业大学计算机科学与技术学院,天津300387

出　　处：《天津工业大学学报》2023年第4期71-76,共6页Journal of Tiangong University

基　　金：国家自然科学基金资助项目(61771340);天津市自然科学基金资助项目(18JCYBJC15300)。

摘　　要：为了准确而高效地识别出布匹各种疵点的种类,采用改进的EfficientDet算法进行布匹疵点识别。首先采取改进的Ostu阈值分割算法进行特征边缘的检测,采用非极大值抑制方法对边缘进行筛选,确定候选区域;然后采用筛选器对候选区域的疵点进行识别和分类,其中筛选器采用改进的EfficientDet算法。改进的EfficientDet算法与其他优秀的目标检测算法以及原算法进行了比较。结果表明,改进的Ostu分割算法相较于传统算法不仅可以在更多的布匹图像中更准确地识别疵点区域,而且抑制了假边缘现象;该模型规模是几种算法中最小的,识别准确率达到94%,高于目前最优算法4个百分点。In order to identify all kinds of fabric defects accurately and efficiently,the improved EfficientDet algorithm is used for fabric defects recognition.The improved Ostu threshold segmentation algorithm is adopted to detect feature edges,and non-maximum suppression is used to screen edges to determine candidate regions.Then a filter is used to identify and classify the defect in the candidate area.The filter uses an improved EfficientDet algorithm.The improved EfficientDet algorithm is compared with other excellent target detection algorithms and the original algorithm.The experimental results show that the improved Ostu segmentation algorithm can accurately identify the defect areas in more fabric images and suppress the false edge phenomenon compared with the traditional algorithm.The model size is the smallest among the centralized algorithms,and the recognition accuracy was 94%,beating the best algorithm by 4 percentage points.

关 键 词：疵点检测 迁移学习 OSTU 目标检测 EfficientDet 

分 类 号：TS101.97[轻工技术与工程—纺织工程] TP391.41[轻工技术与工程—纺织科学与工程]