An Algorithm based on Coherence Reduced Stagewise K-Singular Value Dictionary Learning

Despite rapid advancements in sonar technology, accurately classifying underwater targets remains a significant challenge in the field of underwater signal processing. To address this, this paper introduces a novel methodology that employs dictionary learning algorithms to enhance the expressiveness of dictionaries and align their constituent atoms more closely with the intrinsic structure of signals. Specifically, we focus on underwater target recognition and classification using feature extraction methods based on dictionary learning within a bistatic sonar system. Our primary goal is to develop a robust approach for identifying and categorizing underwater objects, leveraging the unique capabilities of bistatic sonar and the adaptability of dictionary learning techniques. By customizing the dictionary to fit the characteristics of the received signals, our method promises improved accuracy and efficiency in target classification, particularly in complex underwater environments where traditional methods may struggle. Simulation results demonstrate that our target recognition method, based on the Coherence Reduced Stagewise K-Singular Value Decomposition (CRSK-SVD) algorithm, exhibits strong noise resilience and superior overall performance compared to conventional techniques, thereby offering a promising solution for enhancing the reliability of underwater target recognition systems.