Improving performance: A collaborative strategy for the multi-data fusion of electronic nose and hyperspectral to track the quality difference of rice

Although multi-sensor system can obtain the comprehensive information of detected object from different information sources, the direct fusion of multi-data contains a lot of redundant information, which will reduce the detection accuracy. In this work, a collaborative strategy was proposed to track the quality difference of rice, it combined the deep learning and machine learning theory to improve the detection performance of fusion system. Firstly, the quality information of rice was collected based on the electronic nose (e-nose) and hyperspectral imaging system. Secondly, a new structure of convolutional neural network (CNN) was designed to extract the features of fusion data based on the convolution and pooling processes. Finally, a novel global extension extreme learning machine (GE-ELM) was proposed, which combined the dragging factor and global identification coefficients to expand and balance the differences between classes, thereby improving the identification ability and enhancing the stability. Compared with the traditional feature mining and recognition methods, CNN extracted the fusion features effectively, an excellent classification performance of 98.07 % was obtained based on the GE ELM. In conclusion, CNN-GE-ELM was demonstrated as an effective analytical technique to improve the detection performance of fusion system and achieve the high-precision recognition for the quality difference of rice.

Automated summary

The study presents a collaborative strategy to track the quality difference of rice, combining deep learning and machine learning theories to improve detection performance. By collecting quality information of rice and designing a new CNN structure, along with the GE-ELM model, the fusion system's performance and recognition ability were effectively enhanced.