Synthetic Aperture Radar (SAR), owing to its strong capability to operate under adverse weather and illumination conditions, is considered a vital tool for maritime monitoring and surveillance. However, accurate identification of authorized and unauthorized vessels in SAR imagery remains a fundamental challenge due to the presence of speckle noise, complex texture structures, and high variability in the target backscattering characteristics.In this study, a novel approach is proposed to enhance vessel detection in SAR data, with a primary focus on extracting discriminative and robust features from the images. Within this framework, the SOD_TS algorithm is employed to extract salient regions and generate effective feature representations. Subsequently, the extracted features are utilized for classifying authorized and unauthorized vessels using several machine learning classifiers, including Support Vector Machines (SVM), Naïve Bayes, Logistic Regression, Random Forest, Decision Tree, K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), Ensemble Learning, and Quadratic Discriminant Analysis (QDA).This hybrid framework not only improves the generalization capability of the model but also leads to a significant enhancement in accuracy and sensitivity metrics. Experimental results obtained from evaluations on real SAR datasets demonstrate that the proposed method outperforms benchmark approaches such as CNN and YOLO, achieving an accuracy of 0.88, sensitivity of 1.00, F1-score of 0.89, and specificity of 0.99.Overall, this research provides an efficient and generalizable framework for maritime target detection in SAR imagery and is expected to serve as a reliable and practical solution for maritime surveillance applications.

Keywords