Interleaved attention convolutional compression network: An effective data mining method for the fusion system of gas sensor and hyperspectral

Abstract The combination of multi-instrument can comprehensively show the overall attributes of sample from different information sources. However, multi-sensor data fusion brings some redundant information and reduces the recognition accuracy. Therefore, a new deep learning model, namely interleaved attention convolutional compression network (IACCN), is proposed to realize the identification of rice quality in six storage periods under different storage conditions. Firstly, electronic nose (e-nose) and hyperspectral technology are used to get the gas information and spectral information. Secondly, the interleaved attention convolution block (IACB) is proposed in the IACCN to realize the information interaction between the e-nose and hyperspectral data, improve the parameter utilization and extract important features. Finally, knowledge distillation (KD) is introduced to improve the detection performance and stability of the model. Compared with other deep learning methods, IACCN shows a better classification performance and good stability. In conclusion, IACCN is an effective data mining method to improve the classification ability of the fusion system and provides the technology to monitor the rice quality.