Change detection (CD) in urban and natural areas using Very High-Resolution (VHR) satellite images is essential for damage assessment, urban expansion, and environmental analysis. Traditional supervised machine learning techniques face challenges due to urban environments’ spatial and spectral complexity and the difficulty in obtaining extensive training data. This paper introduces an unsupervised CD method for MAXAR VHR images, which addresses these challenges by eliminating the need for prior knowledge. Our approach integrates Multi-Resolution Segmentation (MRS) and the Gas Neural Network (Gas-NN) algorithm to enhance feature extraction and selection. We extract textural and spectral features from pre- and post-event images, using correlation analysis to identify and retain features with high discriminant capability. The Interquartile Range (IQR) method identifies and removes outlier data, thereby improving data quality. The difference map generated from these features is segmented using MRS, with segments represented by their mean pixel values. These segments are then clustered using the Gas-NN algorithm, where the cluster with the highest center values is identified as the changed cluster. Our method achieves an overall accuracy of 97.68% based on ground truth data, demonstrating its effectiveness in automatic CD without extensive training data. This approach shows significant potential for applications in damage assessment, urban expansion, and environmental analysis, marking an advancement in Earth observation and remote sensing.