Application of Hyperspectral Imaging Technologies for Early Detection of Crown Rot Disease in Wheat Under Controlled Environment

Abstract

Crown rot disease is one of the most important stubble-soil fungal diseases faced by the cereal industry globally, causing failure of grain establishment which can lead to significant yield loss. Screening crown rot diseased crops are one of the key elements to manage crown rot, especially in breeding. A key challenge for screening crown rot is there are no clear visible symptoms on upper stems and leaves at early growth stages. The current manual screening method is time-consuming and labour-intensive, requiring pathologists to take detailed observations of crown and root regions of numerous plants in short term for disease inspection. While hyperspectral imaging technologies have been successfully applied to improve screening for some plant health indicators and diseases, they have yet to be successfully used for crown rot disease detection. In this research, we studied hyperspectral imaging technologies to detect crown rot disease at an early growth stage in a high-throughput, accurate, economic and non-destructive manner. Four common Australian commercial wheat varieties with different resistance levels were chosen for study (Aurora, Yitpi, Emu Rock and Trojan). Three different hyperspectral cameras in different wavelength ranges were tested from both side-view and top-view. Four types of input data were tested for support vector machine classification, one was original hyperspectral data and the other three were produced by different hyperspectral image transformation techniques (hyper-hue, standard normal variate, and principal component analysis). The experimental results showed that hyperspectral imaging technologies can be successfully applied to distinguish infected from healthy plants in a greenhouse approximately 30 days after infection, and hyper-hue and standard normal variate data transformation methods provided a high F1 score above 0.75. The results provide a critical first step toward developing a hyperspectral imaging system for crown rot detection to enable the management of crown rot disease and the breeding of resistant varieties.

Keywords: Crown rot, Hyperspectral imaging technologies, machine learning, plant phenotyping, Computer Vision