Data and Code for: Exploring the Relationship Between Cluster Validity Indices and Classification Accuracy in UAV Data-Based Wetland Mapping

Dataset Description: This dataset contains the original UAV imagery and classification code supporting the research article titled "Exploring the Relationship Between Cluster Validity Indices and Classification Accuracy in UAV Data-Based Wetland Mapping".Abstract: In automatic classification frameworks that generate pseudo-labels through clustering, the relationship between Cluster Validity Indices (CVIs) and downstream supervised classification performance remains insufficiently investigated. This study examines this relationship using multispectral UAV imagery acquired over the Bosten Lake wetland in Xinjiang during autumn and winter. An automated classification framework is developed that integrates clustering-based pseudo-label generation with Random Forest (RF) classification. Within this framework, unsupervised clustering first produces pseudo-labeled samples, which subsequently train an RF classifier to derive wetland land-cover maps. The results indicate that the kmeans_rcpp algorithm achieves superior performance in spectrally complex wetland environments. The high-quality pseudo-labels generated by this algorithm substantially enhance RF classification accuracy. External validity indices, including entropy and recall, exhibit positive correlations with downstream classification performance. This approach provides an efficient and reliable solution for fine-scale wetland mapping based on UAV imagery, particularly under conditions of limited labeled samples, and highlights the practical significance of CVIs in automated remote sensing classification.