Improved faster R-CNN for fabric defect detection based on Gabor filter with Genetic Algorithm optimization

Abstract

Fabric defect detection plays a crucial role in fabric inspection and quality control. Convolutional neural networks (CNNs)-based model has been proved successful in various defect inspection applications. However, the sophisticated background texture is still a challenging task for fabric defect detection. To address the texture interference problem, taking advantage of Gabor filter in frequency analysis, we improved the Faster Region-based Convolutional Neural Network (Faster R-CNN) model by embedding Gabor kernels into Faster R-CNN, termed the Genetic Algorithm Gabor Faster R-CNN (Faster GG R-CNN); in addition, a two-stage training method based on Genetic Algorithm (GA) and back-propagation was designed to train the new Faster GG R-CNN model; finally, extensive experimental validations were conducted to evaluate the proposed model. The experimental results show that the proposed Faster GG R-CNN model outperforms the typical Faster R-CNN model in terms of accuracy. The proposed method’ mean average precision (mAP) is 94.57%, compared to 78.98% with the Faster R-CNN.