Tomato Leaf Disease Detection through Machine Learning based Parallel Convolutional Neural Networks

Agriculture is the primary economic activity in a significant number of nations. Yet, plant-related illnesses are the most difficult obstacle for farmers to overcome. The leaf is the portion of the plant that is noticed the most, but the process of segmenting an image of a sick lesion on a leaf suffers from challenges related to uneven lighting and a crowded, complicated natural environment. The combination of colour balancing and superpixel is the foundation of the approach that has been suggested as a solution to these issues. To begin, the picture that is being read in will be converted into a color-balanced image so that the effects of uneven lighting may be removed. Second, using the superpixel operation, compact areas are produced from the modified picture. An experimentally calculated threshold is then applied to the Histogram of Gradients (HOG) and colour channels of the superpixel to separate the undesirable backdrop from the image of the leaf. In order to identify a sick or contaminated picture, K-means clustering is employed. Pyramid of HOG (PHOG), an expanded version of HOG, together with Grey Level Co-occurrence Matrix (GLCM) characteristics are used to depict a sick infected portion of the body. In the end, many disease classifiers are evaluated against one another, and Random Forest (RF) is chosen as the best option. An experiment is carried out to verify the effectiveness of the proposed approach’s accuracy using natural photographs obtained from a variety of farms, images downloaded from the internet, and the Plant-Village colour dataset. This is done so that the resilience of the method can be evaluated.