RGA-Net: An Effective Deep Learning Method Combined With an Electronic Nose System to Identify the Pork Quality

The recognition of gas information based on deep learning methods faces the disadvantages of poor classification stability, low accuracy, and feature degradation. In this work, a residual gas attention network (RGA-Net) is proposed and combined with an electronic nose (e-nose) system to achieve pork quality identification. First, based on the practical engineering application requirements, gas information on pork of different qualities is obtained. Second, considering the data characteristics of gas information, the gas attention module (GAM) is proposed to adaptively focus on the important gas features that improve classification performance. Third, a residual dense block (RDB) is introduced to avoid the gas feature degradation caused by deep networks. Finally, combining RDB and GAM, RGA-Net is proposed to classify gas information from pork of different qualities effectively. The results show that RGA-Net achieves the best classification performance in ablation experiments and compared with other gas information classification methods, with a classification accuracy of 97.75%, a precision of 97.91%, and a recall of 98.39%.