Abstract
<jats:p>Unsupervised clustering remains an essential technique for remote sensing image analysis because ground-truth labels are often unavailable or expensive to obtain. Conventional clustering algorithms, such as K-means and ISODATA, generally rely on distance-based similarity measures, require empirical parameter tuning or predefined numbers of clusters, and become less effective in high-dimensional spectral spaces. Moreover, spectral signatures acquired under different imaging conditions often vary in amplitude while preserving consistent local spectral structures. Unsupervised Wavelet-Feature Greedy Clustering Algorithm (WFGCA) is proposed for multispectral and hyperspectral remote sensing image classification, where multiscale wavelet transforms are employed to identify abrupt spectral structures, including local maxima, minima, and zero-crossing features, which provide robust descriptors of material characteristics. A wavelet-feature correlation coefficient is introduced to measure spectral similarity according to the positional consistency of these characteristic features rather than their amplitudes. Representative cluster centers are then selected using a greedy strategy that minimizes mutual correlation, automatically determining the number of clusters without prior knowledge of class distributions. The proposed method was evaluated using Landsat Thematic Mapper (TM) multispectral imagery and AVIRIS hyperspectral imagery. Parameter analyses demonstrate that the sampling interval, wavelet-feature threshold, orthogonality threshold, and wavelet decomposition scale effectively control clustering granularity and computational efficiency. Experimental results show that WFGCA produces homogeneous clustering results with well-preserved land-cover boundaries while substantially reducing computational cost through spatial sampling and compact wavelet-feature representations. The algorithm performs effectively on both multispectral and hyperspectral datasets without requiring iterative optimization or a predefined number of clusters. By exploiting stable multiscale spectral structures instead of absolute spectral amplitudes, WFGCA provides an efficient and robust unsupervised clustering framework for remote sensing image classification. The proposed wavelet-feature representation, correlation measure, and greedy cluster-center selection improve clustering adaptability, reduce redundancy among representative spectra, and offer flexible control over clustering resolution, making the method suitable for large-scale multispectral and hyperspectral remote sensing applications.</jats:p>