Integrating Efficient Channel Attention Plus into ConvNeXt Network for Remote Sensing Image Scene Classification

Remote sensing image scene classification can play a crucial role in land management. It can extract meaningful information from Earth observation data, yet it continues to face challenges due to complex backgrounds and high intra-class variability. To overcome these challenges, this study proposes an enhanced deep learning framework by integrating Efficient Channel Attention Plus (ECA+) into the ConvNeXt network architecture, aiming to improve feature representation and classification accuracy without adding significant computational cost. Renowned for robust feature extraction capabilities, ConvNeXt significantly benefits from ECA+ as it adaptively recalibrates channel-wise features by prioritizing informative channels and suppressing less relevant ones. The effectiveness of the proposed method was evaluated using the EuroSAT dataset, a widely used benchmark for remote sensing image classification. Experimental results demonstrated that ConvNeXtV1-Nano integrated with ECA+ achieved the highest classification accuracy, weighted precision, weighted recall, and weighted F1-score of 85.56%, 85.82%, 85.56%, and 85.35% respectively, significantly outperforming the base models. These findings highlight the substantial improvement offered by attention mechanisms such as ECA+ in optimizing deep neural networks for more accurate and reliable remote sensing image analysis.