UAV Imagery and Code for Wetland Mapping via Clustering and Random Forest

Autumn and winter multispectral UAV imagery from the Bosten Lake wetland in Xinjiang supports research on automated wetland land-cover mapping. The dataset includes code for an automated classification framework that uses unsupervised clustering to generate pseudo-labels for training a Random Forest classifier. This approach is designed for fine-scale mapping in spectrally complex environments, particularly when labeled samples are limited.

Use Cases

• Evaluating the relationship between cluster validity indices and downstream supervised classification accuracy
• Training Random Forest classifiers for land-cover mapping using clustering-generated pseudo-labels
• Comparing clustering algorithm performance in spectrally complex wetland environments
• Conducting fine-scale wetland mapping with multispectral UAV imagery under limited labeled sample conditions

Strengths

• Includes original UAV imagery and full classification code supporting a published research methodology
• Released under a permissive CC-BY-4.0 license, facilitating reuse and modification
• Focuses on a specific, complex environment (Bosten Lake wetland) across two seasons

Limitations

• Column names and row counts are not specified, limiting precise understanding of data structure
• The dataset size (74,773,255 bytes) is provided, but the composition of files within the ZIP is unclear

Provenance

Source

Jialu Zhang, supporting the research article 'Exploring the Relationship Between Cluster Validity Indices and Classification Accuracy in UAV Data-Based Wetland Mapping'

Collection Method

Multispectral UAV imagery acquisition and subsequent processing through an automated clustering and classification framework

Time Range

Autumn and winter

Freshness

Last updated March 19, 2026

Geography

Bosten Lake wetland, Xinjiang