Knowledge Pre-Trained CNN-Based Tensor Subspace Learning for Tomato Leaf Diseases Detection

Abstract

The early identification of plant diseases is essential for mitigating crop damage and promoting robust agricultural output. By implementing effective disease management strategies, particularly for crops like tomatoes, agricultural yield and sustainability can be greatly improved. This paper introduces HOWSVD-TEDA, an innovative tensor subspace learning technique designed for the detection and classification of diseases in tomato leaves. The approach utilizes advanced pre-trained Convolutional Neural Networks (CNNs) integrated with Higher-Order Whitened Singular Value Decomposition (HOWSVD) and Tensor Exponential Discriminant Analysis (TEDA) to capitalize on the multidimensional representation of data. Extensive testing on the PlantVillage and Taiwan datasets reveals that HOWSVD-TEDA surpasses existing methods, achieving notable accuracy rates of 98.51% and 89.49%, respectively. This advancement represents a significant improvement in the precision and effectiveness of tools for diagnosing tomato leaf diseases.