Application Of Electronic Nose And Hyperspectral In Rice Quality Monitoring And Origin Identification

The quality of rice will change with the prolongation of storage time.If the storage conditions are not suitable,especially if the humidity is too high,the rice will even be mildewed,seriously endangering the health of the body.The quality of rice varies greatly depending on the place of origin.There is a phenomenon that unscrupulous traders label low-quality rice with a high-quality place of origin.According to the needs of rice quality monitoring and origin identification,the work uses electronic nose and hyperspectral technology to obtain rice odor and spectral information.Two kinds of rice,Dao Huaxiang and Yuan Lixiang,are selected as the research objects to study the changes of rice quality under different humidity conditions.The origin identification research is carried out on the rice brands of rice flower fragrance in 7 farms around Jilin.Aiming at the problems of complex internal structure of odor information and spectral information,strong correlation between data,limited amount of data and redundant information,in the work,two data processing methods,Interleaved Attention Convolutional Compression Network(IACCN)and Ultra-lightweight Dynamic Attention Network(ULDAN),are proposed to effectively mine deep features,reduce the complexity of the model and improve the accuracy and performance stability of classification and recognition,so as to realize the monitoring of rice quality under different storage conditions and the identification of the origin of Dao Huaxiang from 7 farms with similar geographical locations in Jilin.The main conclusions are as follows:(1)Based on the odor and spectral fusion data,the rice quality identification model with different storage condition was established.Both IACCN and ULDAN achieved good results.For Dao Huaxiang,ULDAN achieved the highest classification accuracy of 98.5% at 75% RH,and for Yuan Lixiang,IACCN achieved the highest classification accuracy of 99.1% at 75%RH.Through the radar map of the electronic nose sensor response,it can be concluded that when rice is stored in different humidity and different storage time,its sensor response value is different,and the W5 S and W1 W sensors have more obvious responses.The results showed that both the electronic nose and hyperspectral technologies can realize the quality identification of rice under different storage humidity,and the fusion dataset had the highest classification accuracy,followed by the electronic nose dataset and the hyperspectral dataset with the lowest classification accuracy.(2)Based on odor and spectral fusion data,the identification model of Dao Huaxiang origin of seven farms in Jilin region with similar geographical locations was established.The classification accuracy of IACCN was 80.16%,and the ULDAN had the highest classification accuracy of 85.50% and the best performance stability.The results showed that the classification result of the fusion dataset was much larger than that of the single dataset.Taking ULDAN as an example,the classification accuracy of the fusion dataset was 15.00%better than that of the electronic nose dataset and 16.50% better than that of the hyperspectral dataset.