PEANUT CROP PEST DETECTION AND CLASSIFICATION USING MACHINE LEARNING, CONVOLUTIONAL FUZZY LOGIC, AND EVITA

Abstract

Image recognition and classification have both benefited greatly from the fast development of Vision Transformer (ViT) techniques. Specifically for the purpose of identifying, segmenting, and classifying pests, this study presents an Enhanced Vision Transformer Architecture (EViTA). With EViTA, we want to increase the accuracy of pest image prediction by capitalising on ViT's advantages over CNNs. Among the preprocessing methods used in this approach are Moth Flame Optimisation (MFO) for normalising and flattening images, and a dual-layer transformer encoder for integrating pest picture segments of different sizes. The effectiveness of EViTA has been shown by extensive studies that used three insect datasets that impact peanut crops. The findings are encouraging. Research into supplementary methods, including as DenseNet, InceptionV3, and Xception TL models, also points to avenues for accuracy gains above 94%. Also, by using the Flask framework, it is possible to create an authentication-ready testing front end that is easy for users to navigate. EViTA introduces a new way of looking for pests, which might greatly improve farming and pest control. The potential for EViTA to perform better on pest detection tasks might be enhanced with further study and optimisation.