<p>Deep learning-based hyperspectral image (HSI) classification methods have achieved promising results. However, two limitations persist: most models focus solely on either spatial or spectral features, failing to fully exploit the rich frequency-domain information in HSIs; besides, approaches based on convolutional neural networks (CNNs) predominantly employ fixed-size convolutional kernels, struggling to extract multi-scale spatial structures, leading to the loss of local details. To address these issues, a multi-feature fusion network based on wavelet transform and multi-scale cross-response (WTMC) is proposed. Specifically, the wavelet transform-CNN branch is proposed to combine frequency domain information and spatial domain information, where the frequency domain information is extracted by two-dimensional discrete wavelet transform. Meanwhile, multi-scale cross-response branch extracts the spectral-spatial information at different scales, and conducts cross-branch interaction of the information to further facilitate information dissemination. Moreover, the multi-scale pyramid pooling is specifically engineered to expand the receptive field and obtain the global context information better. Extensive experiments are conducted on three public HSI datasets including Salinas Scene (SS), Pavia Center (PC) and Houston dataset. The experimental results with overall accuracy (OA) of 99.63% on SS, 99.68% on PC, and 99.12% on Houston using 5% training samples, conclusively demonstrate that the proposed WTMC framework achieves better classification accuracy compared to state-of-the-art methodologies.</p>