Hyperspectral image (HSI) classification presents a significant challenge due to the high-dimensional nature of the data. Although classical Convolutional Neural Networks (CNNs) are effective at extracting spatial features, they often struggle to capture complex spectral relationships and model the nonlinearities inherent in hyperspectral data, including the non-linear reflectance response across spectral bands and higher-order interactions between wavelengths. In this work, to address the limitations of CNNs, we propose a new hybrid CNN and Kolmogorov-Arnold Network (KAN) framework for HSI classification (HiC-KAN) that integrates the spatial feature extraction capabilities of CNNs with the high- dimensional nonlinear representation capabilities of KANs. The proposed HiC-KAN architecture combines a multi-scale 3D CNN for spatial feature extraction with a parallel KAN branch for spectral nonlinearity modeling, integrated through an attention-enhanced fusion mechanism. This dual-branch design effectively captures both spatial and spectral characteristics while emphasizing discriminative features. Experimental evaluations conducted on the WHU_Hi_Longkou and Xuzhou datasets reveal that HiC-KAN consistently delivers outstanding classification performance, achieving overall accuracy (OA) in the range of 98.04% to 98.07%. This superior accuracy underscores the model’s effectiveness in addressing the challenges associated with hyperspectral image classification.