A Novel Iterative Semi-Supervised Learning Framework based on Few-shot Samples for China Coastal Wetland Land Cover Classification Using GF-5 Hyperspectral Imagery

A novel approach is proposed in this study that combines superpixel (SP) segmentation and multi-classifier ensemble learning to address the limited availability of labeled samples in the coastal wetland land cover classification. Firstly, the SP segmentation techniques is employed to partition unknown samples into multiple homogeneous regions, thereby facilitating the effective capture of spatial information pertaining to land cover. Subsequently, a multiclassifier ensemble learning strategy is employed within these regions to process the samples, effectively leading to a reduction in classification errors and an improvement in accuracy. To enhance the performance of semi-supervised learning (SSL), a sample iteration selection metric is introduced, optimizing the training samples based on the consistency of sample types within homogeneous regions and the results obtained from the multi-classifier ensemble, thus enhancing the reliability of pseudo-labels. Additionally, multi-scale SP segmentation is utilized to augment the ensemble strategy for samples, reducing the necessity for hyperparameter adjustments and increasing the automation and reliability of the model. The effectiveness of the proposed approach has been assessed through experiments conducted on Dafeng Natural Reserve hyperspectral images of wetlands in China.