As a non-destructive technology for detection, hyperspectral imaging has potential for phenotyping plant diseases and providing information for plant protection. Currently, most studies based on hyperspectral imaging technology primarily focus on leaf diseases and less on stalk diseases. In this study, a set of automatic detection and diagnosis methods, combined with spectral and image analyses, was proposed for detecting a stalk disease, rice sheath blight (Rhizoctonia solani), which is widely distributed and deleterious to yield and quality of rice. This study proposed a stepwise method for detecting rice sheath blight. The procedure starts from removal of non-plant background using the k-means clustering algorithm. Then the rice anomalous regions are identified by applying Fisher linear discrimination on sensitive bands. Finally, the scabs of rice sheath blight are detected through a newly developed approach called Hyperspectral Feature Profile Scanning-based Scab Detection (HFPSSD). The validation results showed that the proposed method can effectively recognise disease scabs. The overall accuracy reached 98.42% at pixel level and 95.92% at patch level, which outperformed the traditional support vector machines (SVM) algorithm. The proposed method can potentially serve as a tool for high-throughput detection of plant stalk diseases in the field.

作为一种无损检测技术，高光谱成像具有对植物病害进行表型分析和为植物保护提供信息的潜力。目前，大多数基于高光谱成像技术的研究主要集中在叶片病害上，对茎秆病害研究较少。本研究提出了一套结合光谱和图像分析的自动检测和诊断方法，用于检测水稻纹枯病（Rhizoctonia solani）。)，分布广泛，对水稻产量和品质有害。本研究提出了一种检测水稻纹枯病的逐步方法。该过程从使用 k-means 聚类算法去除非植物背景开始。然后通过对敏感带应用Fisher线性判别来识别水稻异常区域。最后，通过新开发的方法检测水稻纹枯病的痂，称为基于高光谱特征轮廓扫描的痂检测 (HFPSSD)。验证结果表明，该方法能有效识别病痂。整体精度在像素级达到 98.42%，在补丁级达到 95.92%，优于传统的支持向量机（SVM）算法。