融合注意力机制和Faster R-CNN的织物疵点检测算法

Fabric defect detection algorithm based on attention mechanism and Faster R-CNN

针对织物疵点纹理多变、类型多样和尺度不一的特点,提出了优化Faster R-CNN疵点检测模型。将基于卷积模块的注意力机制(convolutional block attention module,CBAM)引入经典Faster R-CNN模型中,对建立的6317张包含污渍、破洞、跳花、断经、断纬、缺经、缺纬和并纬等疵点的织物图片样本库进行CBAM的改进模型与原模型对比实验。结果表明:优化后的网络模型能有效提高织物疵点识别的精度和检测速度,模型的平均精度均值和准确率均值分别从77.01%、61.55%提升到78.81%、64.37%;同时,单张图像的平均检测时间也明显缩短。

In view of the characteristics of fabric defects such as various textures,types,and different scales,an optimized faster R-CNN defect detection model was proposed.The convolutional block attention module(CBAM)was introduced into the classic Faster R-CNN model,and the improved CBAM model was compared with the original model by establishing a large sample of 6317 fabric images with defects such as stains,holes,hops,broken warp,broken weft,missing warp,missing weft and parallel weft.The experimental results show that the improved model can effectively enhance the accuracy and speed of fabric defect recognition.The mean average precision and average accuracy of the model increase from 77.01%and 61.55%to 78.81%and 64.37%,respectively.Furthermore,the average detection time of a single image is significantly shortened.