Block-Segmentation-Based Multi-Dimensional Feature Mining and Intelligent Detection Algorithm for SAR Internal Waves

Synthetic Aperture Radar (SAR) serves as a primary tool for observing oceanic internal waves. To address the challenge of high-precision internal wave detection in SAR imagery, this paper proposes an adaptive feature mining and detection algorithm based on block segmentation. First, discarding conventional global feature extraction, the SAR image is partitioned into multiple equally sized blocks. Second, gray-level co-occurrence matrix (GLCM) and Histogram of Oriented Gradient (HOG) algorithms are employed to extract features from each block, which are then concatenated into feature vectors. Next, principal component analysis (PCA) is applied to reduce feature dimensionality within blocks. A PSO optimized K-means clustering algorithm adaptively clusters and merges blocks into internal wave regions and background regions. Finally, the constant false alarm rate (CFAR)-based edge detection algorithm is utilized to extract internal wave edge features, achieving accurate internal wave region identification. Experimental validation using real internal wave SAR images confirms the effectiveness of the proposed method, offering novel insights and directions for SAR internal wave detection.