Bio-Phantom Net: A Phantom Biochemical Field-Based Deep Learning Model for Accurate and Early Plant leaf Disease Detection

Abstract

Plant diseases drastically decrease crop yield and quality, necessitating precise and early detection techniques. This paper presents Bio-PhantomNet, a new deep learning system, which uses biologically interpretable phantom biochemical fields to identify plant diseases in a variety of crops. The images of the leaves are pre-processed and factorized into the latent features of the chlorophyll degradation, lesion growth potential, and moisture/stress gradient, and reduced in size with the help of PCA to create compact feature vectors. These vectors are then fed into Bio-PhantomNet through fully connected layers, optional attention mechanisms, and then it is followed by a softmax classifier making disease predictions. From the Plant Village dataset, which contains 38 classes including apple, corn, grape, potato, tomato, and pepper, experimental tests were conducted on selected crops: pepper, tomato, and potato. The results show that Bio-PhantomNet achieves high accuracy in disease detection (99.97), precise (99.96), recall (99.95) and F1-score (99.95). The model is very robust, interpretable and reliable, and can thus be applied to real-world scenarios in precision agriculture and smart farming.