<p>Remote Sensing Image Scene Classification (RSSC) plays a vital role in a wide range of applications. In recent years, the rapid development of deep learning has led to the widespread adoption of neural networks in RSSC, yielding impressive results. However, many existing approaches improve classification accuracy by increasing model depth or incorporating self-attention mechanisms, which significantly raise resource consumption and computational overhead–ultimately limiting their practical applicability. To overcome these challenges, we propose a lightweight and efficient convolutional neural network, CEFFNet. This model introduces an Efficient Feature Fusion (EFF) module that effectively enhances both global context understanding and local feature extraction, while substantially reducing computational complexity. Additionally, we design a novel Joint Contrastive Classification Loss (JC-Loss) that constrains sample distribution by reducing intra-class variance and enlarging inter-class separation, thereby further boosting the model’s discriminative power and classification performance. Extensive experiments on three public datasets demonstrate that CEFFNet achieves outstanding classification results while maintaining low computational cost and a compact parameter size.</p>