Fusion of non-iterative deep neural network feature extraction with kernel extreme learning machine for plant disease classification

Plant disease identification and detection refer to the process of recognizing and diagnosing diseases that affect plants. This process is crucial for maintaining plant health, maximizing crop yields, and preventing the spread of diseases to neighbouring plants. Various techniques have been designed for plant disease identification with outstanding results. Still, it exhibits a prolonged convergence rate, taking much time to execute because of long iterations, consuming a large chunk of resources, and suffering local minimum problems. All these issues have emerged from the expansion of hidden layers and their parameters. In this work, a novel hybrid approach is proposed using a ResNet-50 based deep neural network integrated with a Kernel Extreme Learning Machine (KELM) classifier for efficient and accurate plant disease classification. Unlike conventional deep learning methods that rely on iterative training and heavy resources, our approach uses a non-iterative, single-pass KELM classifier to significantly reduce computational complexity while maintaining high classification performance. The method extracts deep, discriminative features via ResNet-50 and feeds them into a lightweight KELM for final classification. The performance of the proposed method is matched with the individual performance of ResNet50 and KELM, along with other Deep Learning (DL) frameworks. Experiments are conducted on the PlantVillage database, which has images of 14 different types of plants with 38 classes. The results obtained provide a nominal error rate of 1.05% using a non-iterative approach, which signifies that the ResNet-KELM method surpasses the existing techniques in terms of accuracy and efficiency.