How to Learn More? Exploring Kolmogorov-Arnold Networks for Hyperspectral Image Classification

TL;DR

This work tackles hyperspectral image classification under data and computational constraints by evaluating Kolmogorov-Arnold Networks (KANs) as data-efficient alternatives to CNNs and ViTs. It introduces HybridKAN, a hybrid architecture that stacks 3D, 2D, and 1D KAN modules with edge-wise learnable activations and a spline-based formulation, preceded by PCA-driven channel reduction. Experiments on three QUH UAV-based datasets (Tangdaowan, Qingyun, Pingan) show HybridKAN achieves competitive or superior performance compared to state-of-the-art CNN- and ViT-based models in terms of OA, AA, and Kappa, with t-SNE visualizations revealing clearer class separation and more homogeneous maps. The results indicate KANs offer fast convergence and robust performance for complex HSIs, supporting their potential as a practical alternative for hyperspectral remote sensing tasks. The work also provides public code to facilitate further research and replication.

Abstract

Convolutional Neural Networks (CNNs) and vision transformers (ViTs) have shown excellent capability in complex hyperspectral image (HSI) classification. However, these models require a significant number of training data and are computational resources. On the other hand, modern Multi-Layer Perceptrons (MLPs) have demonstrated great classification capability. These modern MLP-based models require significantly less training data compared to CNNs and ViTs, achieving the state-of-the-art classification accuracy. Recently, Kolmogorov-Arnold Networks (KANs) were proposed as viable alternatives for MLPs. Because of their internal similarity to splines and their external similarity to MLPs, KANs are able to optimize learned features with remarkable accuracy in addition to being able to learn new features. Thus, in this study, we assess the effectiveness of KANs for complex HSI data classification. Moreover, to enhance the HSI classification accuracy obtained by the KANs, we develop and propose a Hybrid architecture utilizing 1D, 2D, and 3D KANs. To demonstrate the effectiveness of the proposed KAN architecture, we conducted extensive experiments on three newly created HSI benchmark datasets: QUH-Pingan, QUH-Tangdaowan, and QUH-Qingyun. The results underscored the competitive or better capability of the developed hybrid KAN-based model across these benchmark datasets over several other CNN- and ViT-based algorithms, including 1D-CNN, 2DCNN, 3D CNN, VGG-16, ResNet-50, EfficientNet, RNN, and ViT. The code are publicly available at (https://github.com/aj1365/HSIConvKAN)

Figures (13)

• Figure 1: The overall architecture of the Kolmogorov-Arnold Networks.
• Figure 2: Pictorial representation of KAN Convolution operation where $x$, and $\Phi$ represent the input sub-patch and B-splines, respectively. The output of $o_{14} = x \circ \Phi$ can be calculated as $\phi_{11}(x_{11})+\phi_{12}(x_{12})+\phi_{13}(x_{13})+\phi_{21}(x_{21})+\phi_{22}(x_{22})+\phi_{23}(x_{23})+\phi_{31}(x_{31})+\phi_{32}(x_{32})+\phi_{33}(x_{33})$.
• Figure 3: The overall architecture of the proposed Hybrid KAN.
• Figure 4: Pictorial view of the QUH-Tangdaowan data benchmark: (a) the annotation of the training samples, (b) the annotation of the validation samples, and (c) the test samples.
• Figure 5: Pictorial view of the QUH-Qingyun data benchmark: (a) the annotation of the training samples, (b) the annotation of the validation samples, and (c) the test samples.