Research on Fabric Defect Detection Based on Deep Fusion DenseNet-SSD Network

Defect detection to control the quality of fabrics is one of the key tasks in the production process of fabrics. Although significant progress has been made in the research of fabric defect detection, while traditional methods are still difficult to cope with complex and variable defect shapes. In order to solve these problems, this paper proposes an adaptive fabric defect detection method based on DenseNet-SSD algorithm to improve the performance of fabric defect detection. This method uses the DenseNet network to replace the backbone network VGG16 in the SSD algorithm, which strengthens the transfer between feature maps, alleviates the problem of gradient disappearance and reduces the number of network parameters. Compared with SSD, it improves network detection accuracy and real-time performance. The accuracy in the test set is 78.6mAP and the detection speed is 61FPS.