Demystifying KAN for Vision Tasks: The RepKAN Approach

TL;DR

RepKAN is a novel neural architecture combining CNN efficiency with KAN interpretability, enabling physically meaningful reasoning in remote sensing image classification, outperforming existing models on key datasets.

Contribution

Introduces RepKAN, a dual-path model integrating CNNs and KANs for interpretable remote sensing classification, advancing explainability and performance.

Findings

Outperforms state-of-the-art models on EuroSAT and NWPU-RESISC45 datasets.

Provides explicit physically interpretable reasoning.

Enables discovery of class-specific spectral fingerprints.

Abstract

Remote sensing image classification is essential for Earth observation, yet standard CNNs and Transformers often function as uninterpretable black-boxes. We propose RepKAN, a novel architecture that integrates the structural efficiency of CNNs with the non-linear representational power of KANs. By utilizing a dual-path design -- Spatial Linear and Spectral Non-linear -- RepKAN enables the autonomous discovery of class-specific spectral fingerprints and physical interaction manifolds. Experimental results on the EuroSAT and NWPU-RESISC45 datasets demonstrate that RepKAN provides explicit physically interpretable reasoning while outperforming state-of-the-art models. These findings indicate that RepKAN holds significant potential to serve as the backbone for future interpretable visual foundation models.