Advancing spaceborne image analysis: genetic algorithm for automated clustering and spectral optimization in multispectral remote sensing
Advances in Space Research, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.asr.2026.05.098
- Dergi Adı: Advances in Space Research
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Artic & Antarctic Regions, Compendex, INSPEC, MEDLINE, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Anahtar Kelimeler: Band selection, Genetic algorithm, Multispectral imagery, Satellite remote sensing, Unsupervised clustering
- Eskişehir Osmangazi Üniversitesi Adresli: Evet
Özet
Unsupervised clustering of multispectral satellite imagery is hindered by high dimensionality, spectral redundancy, and the need for manual specification of cluster count—challenges that increase downlink data volume and delay real-time analysis in bandwidth-constrained Earth observation missions. This study introduces the first fully unsupervised genetic algorithm (GA) that jointly optimizes cluster count and per-cluster spectral band selection in a single pipeline. The proposed method dynamically determines the optimal number of clusters and selects the most informative bands for each cluster, thereby eliminating noisy and redundant channels while preserving classification integrity.The framework was validated on Sentinel-2, Landsat-8, and PlanetScope imagery over a heterogeneous Mediterranean test site (Kumluca, Türkiye) and further tested on an independent site (Gemlik, Türkiye). Using 10 independent sets of 200 stratified random reference points, the GA achieved mean overall accuracies (OA) of 0.88 (Sentinel-2), 0.87 (Landsat-8), and 0.92 (PlanetScope). It outperformed the ISO Clustering (ISODATA) algorithm by 7, 6, and 3 percentage points, respectively. The method also demonstrated competitive or superior performance against two additional strong baselines—K-Means with stacked autoencoders + elbow method and Self-Organizing Maps (SOM)—across all three platforms. Superior results were obtained with Sentinel-2 due to its balanced spatial-spectral resolution. An additional validation on the Gemlik site with Sentinel-2 data yielded a mean OA of 0.91 (best run: 0.95).The scalable, mission-ready framework supports multi-platform data fusion, significantly reduces redundant spectral data prior to downlink, and enables near real-time applications in disaster monitoring, precision agriculture, and global land cover mapping. By automating both cluster determination and per-cluster band selection in a fully unsupervised manner, the proposed GA advances onboard processing capabilities for current and future spaceborne multispectral systems.