Improved Siamese network algorithm design for remote sensing point cloud data classification

Traditional intelligent algorithms have insufficient utilization of spatial structure information, limited multi-scale feature fusion ability, and weak global contextual correlation in small sample classification tasks of remote sensing point cloud data. In response, a research proposes an intelligent algorithm based on an improved Siamese Network, combining meta learning, self attention mechanism, and multi-scale feature fusion strategy. According to the experimental findings, the suggested model's classification accuracy in the 5-way 1-shot and 5-way 5-shot tasks on the ModelNet40 dataset was 78.3% and 85.7%, respectively; On the self built RS PointCloud dataset, the model accuracy was further improved to 81.5% (5-way 1-shot) and 87.6% (5-way 5-shot) by introducing self attention mechanism and multi-scale feature fusion. Overall, the research model has improved the accuracy and robustness of small sample point cloud classification by enhancing feature discriminability and long-range dependency modeling, providing an effective solution for remote sensing intelligent interpretation tasks.