In the joint classification of multi-sensor remote sensing data, such as hyperspectral images (HSI), light detection and ranging (LiDAR), and synthetic aperture radar (SAR), differences in imaging mechanisms and measurement characteristics often lead to insufficient feature extraction and limited collaborative utilization. To address these issues, this paper proposes a cross-sensor frequency-aware fusion network (CFF-Net) that integrates frequency-domain feature fusion with a hyper-connection mechanism. From a frequency-domain perspective, the proposed method characterizes the complementary relationships among multi-source observations, enabling the collaborative modeling and effective fusion of heterogeneous sensor features. In the feature extraction stage, a multi-dimensional feature perception encoder is constructed to jointly model high-frequency local detail information and low-frequency global structural information in parallel, thereby achieving complementary representations of different sensor observation characteristics. In the feature fusion stage, a cross-sensor hyper-fusion module is introduced, combining the hyper-connection mechanism with frequency-based self-attention to realize adaptive alignment and collaborative enhancement of multi-sensor features in the frequency domain, thereby improving the consistency and discriminability of the fused representations. Extensive experiments are conducted on three public multi-sensor remote sensing datasets: MUUFL Gulfport (MU), Houston 2013 (HU), and Augsburg (AU). The results demonstrate that CFF-Net achieves overall accuracy (OA) of 92.23%, 93.51%, and 91.54%, alongside average accuracy (AA) of 92.05%, 93.77%, and 64.81%, respectively, significantly outperforming all compared methods. With only 0.30 M parameters and 60.55 M FLOPs, the proposed method achieves a favorable balance between classification performance and computational efficiency, validating its effectiveness and applicability in multi-sensor remote sensing data collaborative classification.