Convolution Neural Network (CNN) Approach for Classification of Diseased and Healthy Paddy Crop using UAV-based Multispectral Imageries

The conventional way to detect plant defects is tedious and inefficient through human vision. It requires deep knowledge gained through years of experience, ground observations and understanding of the plant. Therefore, the intelligent methods in this research are expected to assist the farmers in identifying whether a region is disease-affected or healthy. The proposed study aims at the image processing technologies for disease identification using different band images acquired through Unmanned Aerial Vehicle (UAV) mounted with a multispectral camera in the paddy domain. The multispectral imageries were obtained at 30 m altitude to detect diseases in a paddy cultivar (MTU1010) affected by grain discolouration disease. The deep learning method of Convolution Neural Network (CNN) with VGG 16 architect was proposed to classify healthy and diseased images. In the image classification process, the following combinations such as (NIR, RED, NDVI) or (NIR, RED_EDGE, NDVI) or (NIR, RED, NDRE) or (NIR, RED_EDGE, NDRE) were used to identify whether an image is healthy or diseased depending upon their training accuracy, validation accuracy, precision, recall, F1 score and Kappa coefficient. The results showed that the combination of (NIR, red, NDVI) and (NIR, red, NDRE) gives the best classification for diseased and healthy identification. The proposed method is expected to reduce the risk of disease spread over the entire field, which may increase the paddy yield.

Keywords: Disease classification, CNN, NDVI, Multispectral imageries, UAV