Research On The Origin And Variety Identification Of Rice Based On Hyperspectral Image

In recent years,public health incidents have been frequent,and food safety has been paid more and more attention.Quality inspection and traceability have become the key links.As the staple food of rice,its quality detection and origin tracing are very important.The existing detection methods are mainly manual discrimination and chemical analysis,which have the problems of slow detection speed and low accuracy.According to the dimension reduction and classification algorithm of hyperspectral image,this paper proposes a fast and nondestructive identification method for rice origin and varieties based on hyperspectral image.The specific work is as follows:(1)In order to solve the problems of strong correlation between spectra and serious information redundancy of hyperspectral images,an optimal index method based on band grouping and superposition and an index method of inter-species separability are proposed.The experimental results show that the optimal index method based on band grouping and superposition reduces the number of band combinations of hyperspectral images to less than1/10000 of the original algorithm,which effectively reduces the calculation amount of the optimal index method when selecting the band of hyperspectral images.(2)Aiming at the problem that RPNet feature extraction does not consider the size relationship of the measured values among pixels in the local area of the image,an improved RPNet is proposed,which combines RPNet and LBP to extract the spatial features of hyperspectral images.The experimental results show that the improved RPNet is used to extract the spatial features of the selected band images,and the space spectrum joint classification is carried out by SVM.Compared with the former improved RPNet combined with SVM classification,the OA and Kappa of the three public data sets are both increased by more than 6%.In the final CGCP Rice,OA and Kappa increased by 3% and 3.46%,respectively.The OA of CGCP Rice8 and CGCP Rice6 was increased by 7.97% and 1.53%,respectively.(3)SVM or RMG algorithm was used to classify the joint features of the extracted space spectrum.Aiming at the problem of time-consuming classification of RMG algorithm,an improved RMG algorithm was proposed,which improved the way of randomly selecting features when constructing images.The classification accuracy of the improved algorithm was less than 1% lower than that of the original algorithm,but the classification time was reduced to half of that of the original algorithm,which effectively improved the classification efficiency of RMG algorithm.The proposed scheme can effectively improve the processing speed and classification accuracy of the collected hyperspectral images of rice,which is of great significance for the application of hyperspectral image classification technology in rice and other grain sorting tasks,and has a certain reference value for the practical application of hyperspectral image detection technology.