Improving Imbalanced Land Cover Classification with K-Means SMOTE: Detecting and Oversampling Distinctive Minority Spectral Signatures

Land cover maps are a critical tool to support informed policy development, planning, and resource management decisions. With significant upsides, the automatic production of Use/Land Cover has been topic interest for remote sensing community several years, but it is still fraught with technical challenges. One such challenge imbalanced nature most remotely sensed data. The asymmetric class distribution impacts negatively performance classifiers adds new source error these maps. In this paper, we address learning problem, by using K-means Synthetic Minority Oversampling Technique (SMOTE) as an improved oversampling algorithm. SMOTE improves quality newly created artificial data addressing both between-class imbalance, traditional oversamplers do, also within-class avoiding generation noisy while effectively overcoming imbalance. compared three popular methods (Random Oversampling, Borderline-SMOTE) seven benchmark datasets, (Logistic Regression, K-Nearest Neighbors Random Forest Classifier) evaluation metrics five-fold cross-validation approach different initialization seeds. statistical analysis results show that proposed method consistently outperforms remaining producing higher land classifications. These suggest LULC can benefit significantly from use more sophisticated spectral signatures same vary according geographical distribution.