Modified Combined Mobilenet for Tomato Leaf Disease Classification

Abstract

The spread of plant diseases presents a major obstacle to ensuring global food security, as it can significantly reduce agricultural productivity and, in severe cases, lead to complete crop failure. Precisely recognizing plant diseases is essential for implementing effective control strategies, enabling timely intervention to minimize damage and ensure sustainable food production. In this paper, we introduce an improved variant of the combined Mobilenet model, integrating a more efficient convolutional layer within the LRCA Fused MBC layer to improve performance. To further optimize the model, we apply architectural modifications aimed at reducing training time while maintaining high classification accuracy. Transfer learning is employed to leverage pre-trained knowledge, ensuring effective feature extraction and faster convergence. Training and evaluation are conducted using the Kaggle Tomato Plant Disease dataset. Experimental results demonstrate that our modified architecture surpasses the original model, achieving a testing accuracy of 98.6% while significantly reducing computational costs.