Integration of spectroscopy and image for identifying fusarium damage in wheat kernels

Authors: ZHANG, DONGYAN - Anhui Agricultural University; CHEN, GAO - Anhui Agricultural University; Zhang, Huihui; GU, CHUNYAN - Anhui Agricultural University; WANG, QIAN - Anhui Agricultural University; CHEN, YU - Anhui Agricultural University

Submitted to: Spectrochimica Acta
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/5/2020
Publication Date: 4/6/2020
Citation: Zhang, D., Chen, G., Zhang, H., Gu, C., Wang, Q., Chen, Y. 2020. Comprehensive analysis of fusarium head blight in wheat kernel using hyperspectral spectrum and image. Spectrochimica Acta. 236. https://doi.org/10.1016/j.saa.2020.118344.
DOI: https://doi.org/10.1016/j.saa.2020.118344

Interpretive Summary: Due to the health risk associated with toxins produced by Fusarium head blight , it is urgent to develop methods for its detection in wheat kernels. In this study, a total of 810 wheat kernel samples were selected and divided into three grades: healthy, mildly infected and severely infected, according to their degree of infection. Hyperspectral images (400-1000nm) were acquired, pre-processed, and then the optimal wavelengths were selected using three methods: principal component analysis (PCA), successive projection algorithm (SPA) and random forest (RF) algorithm. The image features corresponding to the optimal wavelengths were extracted, and then both spectral and image features were used as inputs to three models used for classification: support vector(SVM), random forest (RF) and naïve Bayes (NB). The best results were obtained by using the SPA-RF algorithm to select the wavelengths and its image features. The classification accuracy of calibration, verification, and prediction prediction sets for SPA-RF are 100.00%, 98.24% and 96.44%, respectively. This study shows that the hyperspectral imaging system has the potential to identify wheat kernel damaged by Fusarium.

Technical Abstract: Fusarium produces toxins in the process of infecting wheat kernels, which can cause great harm to human and animal health. In order to identify infected wheat kernels, this paper investigated the use of hyperspectral imaging (HSI) to identify wheat kernel damaged by Fusarium. In this study, a total of 810 wheat kernel samples were selected and divided into three grades: healthy, mildly infected and severely infected according to their degree of infection. Hyperspectral images containing spectral and image information were acquired over a wavelength range of 400-1000 nm, pre-processed with spectral data, and then the optimal wavelengths were selected using principal component analysis (PCA), successive projection algorithm (SPA) and random forest (RF). The image features of the image corresponding to the optimal wavelengths were extracted, and then the spectral features and image features were fused and used as input of support vector machine (SVM), random forest (RF) and naive Bayes (NB). The results show that the best results were obtained by using the SPA-RF algorithm to select the wavelengths and its image features. The classification accuracy of calibration set, verification set, and prediction sets for SPA-RF are 100.00%, 98.24% and 96.44%, respectively. This study shows that the hyperspectral imaging system has the potential to identify wheat kernel damaged by Fusarium, and the method in this paper improves the utilization of spectral information and image information, which is better than a single spectral model and image model.