Ensemble of Tversky-Indexed Graph Neural Network and CNN for Plant Leaf Disease Prediction

Plant leaf diseases significantly impact agricultural productivity, leading to substantial yield losses and posing a threat to global food security. Early detection and accurate diagnosis are critical for minimizing crop damage and preventing disease spread. However, conventional machine learning and deep learning models often fail to achieve high accuracy and rapid disease classification, limiting their effectiveness in real-time scenarios. To overcome these limitations, an Ensemble of Tversky Indexive Projected Graph Neural Network with Gaussian Distributed Convolutional Neural Network (ETIPGNN-GDCNN) is proposed. This model integrates IoT technology to capture large volumes of plant leaf images, enabling real-time monitoring and disease prediction. The ETIPGNN-GDCNN model consists of two primary components. The Tversky Indexive Projected Graph Neural Network (TIPGNN) transforms plant leaf images into graphs, where pixels act as nodes. It applies Gaussian distributed bilateral filtering to eliminate noise and enhance image quality. The local pooling layer segments Regions of Interest (ROIs) using the Tversky similarity function, simplifying the classification process. Meanwhile, the Gaussian Distributed Convolutional Neural Network (GDCNN) extracts essential features using convolutional and max-pooling layers. Disease classification is performed using Sokal-Michener's Simple Matching technique. The outputs from GNN and CNN are fused using the Nesterov Accelerated Gradient method to minimize classification loss and improve prediction accuracy. Performance evaluation using metrics such as peak signal-to-noise ratio, precision, recall, and prediction time demonstrates that the proposed ETIPGNN-GDCNN model achieves higher accuracy and faster disease detection, making it a promising solution for smart agriculture applications.