Plant Leaf Disease Classification Using Multiscale Pyramid Autoencoder with Kolmogorov–Arnold Network

Abstract

The plant leaf diseases classification is important for crop yield management and market value. Effective prevention and treatment depend on the rapid and accurate identification of leaf diseases and assessment of their severity. However, hyperspectral images capture detailed spectral information across hundreds of narrow bands. The full-spectrum data are high-dimensional, which increases redundancy and computational complexity. Subtle nutrient-induced biochemical changes in plant leaves produce weak spectral variations that fail to generate strong discriminative features for classification tasks. Hence, this research proposes a Multiscale Pyramid Autoencoder with a Kolmogorov–Arnold Network (MPA-KAN) for plant leaf disease classification. An autoencoder components learn a compressed representation of high-dimensional hyperspectral data, reducing dimensionality by effectively eliminating redundant information across multiple spectral bands. The KAN significantly captures both spectral and spatial features while eliminating irrelevant information, thereby enhancing the model efficiency. The proposed MPA-KAN model achieves 98.99% accuracy in the hyperspectral image and 99.98% accuracy in the PlantVillage dataset when compared with existing methods.