BMA-CenterNet based multi-leaf multi-disease classification and invasive plant identification framework using cdeboar-50

Abstract

The early detection of plant Leaf Disease (LD) is crucial for maintaining the crop’s health. Prevailing works overlooked the ageing factor, nutrition factor, water content, chlorophyll, fungi, virus, and bacteria for multiple LD prediction. Therefore, a novel multi-leaf, multi-disease classification and invasive plant identification is proposed. The methodology starts with Gaussian Filter (GF) and Luminance and Chrominance (L&C)-based pre-processing, followed by Leaf Area Measurement (LAM) using the Bhattacharyya Distance-Based Triangulation Method (BDBTM). Invasive plants are then identified and removed using the Cauchy Distributed EBola Optimization Algorithm ResNet-50 (CDEBOAR-50). Further, the edges, veins, and diseased leaf parts are segmented by using Fractal Dimensions (FD). The 3-dimensional-based K-Means Clustering (d3-KMC) differentiates healthy leaves from diseased leaves. Lastly, multiple LDs are classified using Beta-Mish Activated CenterNet (BMA-CenterNet). The proposed model attained an accuracy of 99.77 % and a Matthew’s Correlation Coefficient (MCC) of 0.963986476, outperforming the state-of-the-art approaches and enhancing the smart farming system.