Efficient Remote Sensing Image Classification using the Novel STConvNeXt Convolutional Network

Abstract Remote sensing image classification poses significant challenges due to complex spatial organization, high inter-class similarity, and large intra-class variance. To address these issues, we propose STConvNeXt, a novel pure convolutional neural network specifically tailored for efficient remote sensing image classification. STConvNeXt integrates a split-based mobile convolutional module, a tree structure, and a fast pyramid pooling module to achieve residual connectivity. Additionally, we introduce a threshold loss function to stabilize model training and improve classification accuracy. Comprehensive experiments on multiple remote sensing datasets demonstrate that STConvNeXt achieves a 56.49% reduction in parameters and a 49.89% decrease in computational load compared to ConvNeXt, while maintaining state-of-the-art classification accuracy. Our results highlight the effectiveness of STConvNeXt in extracting robust features from remote sensing images, advancing the frontiers of deep learning-based remote sensing analysis.