Yarn target detection of a braiding machine based on the YOLO algorithm

Abstract

Braiding machines occupy an important position in the textile industry. Aiming at the characteristics of high real-time requirements for yarn target detection in braiding machines, small yarn change curvature, and large background interference, based on the YOLOv7 algorithm model, the lightweight convolution GSConv and VoVGSCSP modules are used to replace the ELAN-H module in the YOLOv7 algorithm to reduce the complexity of the model and improve the detection speed. In order to solve the problems of confusing detection target categories and poor detection effect of targets with small curvature change, a new bounding box loss function, wise intersection over union loss, is introduced to solve the imbalance of sample quality and improve the robustness and generalization ability of the model. The ablation experiment proves that the added modules can be well fused together. The mean average precision, precision, recall, frames per second, and GFLOPs of the improved YOLOv7 are 92.2%, 93.1%, 89.7%, 123.6, and 89.9, respectively.