Quantification of rice spikelet rot disease severity at organ scale with proximal imaging spectroscopy

Spikelet diseases pose severe threats to crop production and crop protection requires timely evaluation of disease severity (DS). However, most studies have only investigated the spikelet diseases within a short period of crop growth. Few have examined the consistency in DS monitoring accuracy across growth stages. This study aimed to investigate the differences in spectral responses among growth stages and to develop a spectral index (SI), rice spikelet rot index (RSRI), for multi-stage monitoring of the rice spikelet rot disease. Proximal hyperspectral images were collected over spikelets with various levels of DS at heading, anthesis, and grain filling stages. The reflectance was related to the DS extracted from concurrent high-resolution RGB images. The proposed RSRI was evaluated for the DS estimation and lesion mapping across growth stages in comparison with existing SIs. The results demonstrated that the spectral responses to DS in the green and near-infrared regions for filling were weaker than those for anthesis, and blue bands were necessary in DS quantification for early infection. The RSRI-based models exhibited the best validation accuracy for heading and the most consistent performance across growth stages as comparison to other SIs (Heading: R² = 0.65; anthesis: R² = 0.84; filling: R² = 0.78). Moreover, RSRI-based DS maps exhibited the best lesion identification for slightly, mildly, and severely infected spikelets. This study suggests that RSRI could be promising in breeding and crop protection as a novel index for DS estimation regardless of the spikelet ripening effect.