Optimal Wavelength Selection Of Hyperspectral Image For Maize Seed Based On Tensor Analysis

Maize,as the most productive crop all over the world,is widely used in food production,industrial raw materials manufacturing and livestock feed processing.The identification of maize seed variety is of great value in reducing the seed mix and ensuring the smooth progress of agricultural production.Hyperspectral image technology has the feature of integrating the feature of image and spectral,which can obtain the image information and spectral information of maize seeds at the same time.Therefore,Hyperspectral image technology is getting more and more attention in maize seed variety recognition,and achieves high recognition accuracy.Sufficient extraction of classification features guarantees the accuracy and robustness of the identification models.Although hyperspectral image technology can obtain image feature and spectral feature of seeds,the spectral feature is the most commonly used in seed variety classification,as a result,people fail to maximize use hyperspectral image technology.Additionally,the hyperspectral images with more wavelengths make the real-time,online application of hyperspectral image seed variety identification difficult.In this paper,a fast,accurate and robust method for nondestructive detection of maize seeds is developed based on the combination of hyperspectral image technology and multi-feature wavelength selection method with tensor analysis.The main research contents include:1.This study aimed to select optimal wavelengths from hyperspectral image data using joint skewness(JS)algorithm,which can be developed in multispectral imaging-based inspection system for the automatic classification of maize seed.The hyperspectral images covering the wavelength range of 400–1,000 nm were acquired for 1632 maize seeds including 17 varieties.The JS algorithm was used to select optimal wavelengths,and the classification models based on least square support vector machine(LS-SVM)were developed using the combined features.The experimental results indicated that the classification model based on JS algorithm yielded the better classification accuracy than that of uninformative viable elimination algorithm(UVE)and successive projections algorithm(SPA).This study provides a feasible approach for the accurate and rapid identification of the hyperspectral image technology available for seed identification.2.Multi-linear discriminant analysis(MLDA),a supervised wavelength selection method,was adopted to study the maize seed purity detection using hyperspectral image.The classification models based on LS-SVM were developed using the feature wavelength sets selected by MLDA and JS,and the recognition accuracy was compared under the same conditions.Experimental results showed that the MLDA wavelength selection method had higher efficiency than the JS wavelength selection method in the same wavelength number,and at the same accuracy.In addition,the MLDA wavelength selection method can obtain less number of wavelengths under the same precision conditions,which was more advantageous to the development of the efficient multispectral image system.3.Hyperspectral image-based variety discrimination of maize seeds with genetic relationship had been studied by using a multi-model strategy and MLDA-based wavelengthselection algorithm.Firstly,the inter class switching model was established for two varieties of maize in the range of 874-1734 nm wavelength,and then subdivided by constructing sub-models in two classes.MLDA-based wavelength selection algorithm was used to improve the speed of detection.Experimental results showed that the multi-model for full-wavelength data and optimal wavelengths both achieved high classification accuracy,and it also has high robustness under different scenes.This study shows that the multi-model coupled with MLDA exhibits high potential in rapid and highly accurate classification of seed varieties.