Machine learning for diagnosis of disease in plants using spectral data

Automating crop disease diagnosis is an important task, particularly for regions with few experts. Most current methods detect disease by analyzing leaf images, particularly for diseases that manifest on the aerial part of the plant. To train a good classifier one requires a huge image dataset and the appropriate methods to extract relevant features from the images that represent the disease unambiguously. Image data also tends to be prone to effects of occlusion that make consistent analysis of the data hard. In this paper we take a look at the use of spectral data collected from leaves of a plant. We analyse spectral data from visibly diseased parts of a leaf as well as parts that are visibly healthy. We employ prototype based classification methods and standard classification models in a three-class classification problem configuration. Results presented show significant improvement in performance when spectral data is used and the possibility of early detection of disease before the crops become visibly symptomatic, which for practical reasons is very important.