Cigarette defect detection algorithm based on attention mechanism and multi-gradient feature fusion

Surface defect detection remains a persistent and challenging task. Aiming at the detection of surface defects in cigarettes, we propose an enhanced YOLOX-S model. Firstly, an improved attention mechanism named MS-GCT (Multi-Spectral Gaussian Context Transformer) is introduced into the model’s backbone to enhance the model’s ability of capturing the global context information within images and improve its comprehension of semantic feature information; secondly, we propose the DMG (Dynamic convolution and MS-GCT) module, and combined with the C2f (CSPLayer with 2 convolutions) module to construct the C2f-DMG module,which is introduced into the model to enhance feature interaction and feature extraction ability, to strengthen long-distance dependency ability of global features; finally, we replace the loss function with SIoU to enhance model performance and accelerate model convergence. To validate the effectiveness of our model, we conduct experiments on both the self-made cigarette dataset and the public dataset. The experimental results indicate that the improved model not only ensures the lightweight of the model, but also boosts the model’s mAP by 2.02, while achieving a detection speed of 73.17 frames−1. Furthermore, the proposed algorithm fulfills the real-time detection requirements for cigarette appearance defects.