FARLut: a two-stage tobacco foreign body detection model incorporating color information and attention mechanism

Accurate detection and rejection of debris in tobacco products play an essential part in ensuring the quality of tobacco products. In recent years, the detection of detritus in the production process has been widely investigated, but there is still room for further improvement in the research of visible light-based tobacco debris detection methods. In this study, we collected visible light images from the tobacco industry production line and constructed a dataset for tobacco debris detection. In addition, a we propose a tobacco debris detection model named the FARLut. The FARLut model preprocesses images obtained from the tobacco production process based on the color and then inputs the processed images into a two-stage target detection algorithm with an attention mechanism for debris detection. The proposed model is verified experimentally. The experimental results show that the FARLut model achieves an average accuracy of 94.91% and a recall rate of 97.20% on the test dataset. Thus, the proposed detection model can effectively identify familiar clutter in tobacco production. The results of this study provide a useful reference for further research on clutter detection in the tobacco production field.