<p>Recent advancements in remote sensing technology have significantly improved image quality, which in turn demands more efficient scene recognition methods. Convolutional neural networks (CNNs) have demonstrated remarkable results but often suffer from high complexity and large parameter counts, limiting their deployment on lightweight devices. This paper introduces LRRN, a lightweight CNN designed for remote sensing image scene recognition. LRRN comprises a Feature Integration Network (FIN) for multi-scale feature extraction and fusion, and a Classification Network Head (CNH) for sparse feature aggregation and classification. The FIN employs a parallel path extraction mechanism to adaptively capture features at various scales, while the CNH utilizes a tandem pooling layer to efficiently aggregate key features. Experiments on public datasets NWPU RESISC-45, EuroSAT, and RSSCN7 demonstrate recognition accuracies of 94.81%, 96.44%, and 83.75%, respectively, with significantly reduced parameter counts and inference times. Furthermore, transfer learning experiments on a self-compiled dataset validate LRRN’s strong generalization capability. Here, we show that LRRN achieves an optimal balance between accuracy and efficiency, making it suitable for deployment on resource-constrained devices. The proposed LRRN and HBUA-NR5 datasets are publicly available at <a href="https://github.com/ysuc/LRRN">https://github.com/ysuc/LRRN</a>.</p>