Improved MobileNetV2-SSDLite for automatic fabric defect detection system based on cloud-edge computing. (30th September 2022)

• Record Type: Journal Article
• Authors: Zhang, Jiaqi; Jing, Junfeng; Lu, Pengwen; Song, Shaojun
• Is Part Of: Measurement. Volume 201(2022)
• Publication Date: 2022-09-30
• Subjects: Edge computing -- Fabric defect detection -- Deep learning -- Image processing -- Convolutional neural network; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.111665
• Languages: English
• ISSNs: 0263-2241

Abstract

Abstract: Fabric defect detection is the important step of ensuring the quality and price of textiles. In order to make the automatic fabric defect detection system used in production sites, a cloud–edge collaborative fabric defect system is proposed. Firstly, real-time defect detection is performed on edge device, the accuracy of small defect detection is ensured by improved MobileNetV2-SSDLite. The channel attention mechanism is introduced in the network to highlight defect features and suppress background noise features. The loss function is redefined by Focal Loss to overcome the imbalance of the number of defects and background candidate boxes. Then, detection result storage and model update are carried out in the cloud. Experiments show that the accuracy of the system is improved while maintaining the faster detection speed, among which, the accuracy of the Camouflage dataset with small defects has increased by 10.03% and the detection speed reaches 14.19FPS on NVIDIA Jeston Nano. Highlights: In the network improvement, the channel attention mechanism extract more refined defect features under the complex background disturbance, improving small defect detection accuracy. And Focal Loss solve the imbalance between defect and background default boxes number in the training process. A lightweight network for defect detection in resource-constrained scenarios is proposed. Detection accuracy is improved for small defects and complex background datasets. Our method balances accuracy and inference time better, providing a new option for the embedded platform. A cloud–edge collaborative fabric defect detection system is proposed. Combining edge devices and the cloud, the detection model is deployed on the edge device for real-time defect detection, data storage and model update at the cloud.