Development of new indices and use of CARS-Ridge algorithm for wheat fusarium head blight detection using in-situ hyperspectral data

Abstract

Wheat Fusarium head blight (FHB) poses a significant threat to wheat quality and yield. However, accurately identifying the disease in wheat ears remains challenging due to limited data and weak hyperspectral signals. This study aimed to address these issues by obtaining effective wheat canopy hyperspectral data over three years of successive experiments. To reduce band redundancy and improve accuracy, nine different models were constructed, and the optimal algorithm was determined. Additionally, two types of new indices, were developed based on the spectral response mechanism of the wheat disease and published vegetation indices. Our study demonstrated that these newly constructed indices outperformed the published vegetation indices in terms of detection capability. By fusing the optimal algorithm with the new indices, a detection accuracy of 91.4% for the disease was achieved, surpassing the current level of wheat FHB detection. The high-accuracy model developed in this study not only provides methodological support for detecting wheat FHB but also serves as a reference for diagnosing diseases in other crops.