Abstract: Objectives To address issues such as susceptibility to environmental noise and blurred target characteristics in underwater sonar images, a recognition method that integrates multi-dimensional features with Gaussian process-optimized least squares support vector machines (LSSVM) is proposed. Methods First, considering the noise characteristics of sonar images, a target enhancement and segmentation algorithm is designed to highlight regions of interest. Subsequently, features are extracted from multiple dimensions, and analysis of variance (ANOVA) along with correlation analysis are applied for feature selection and dimensionality reduction. Finally, Gaussian process optimization (GPO) is introduced to optimize the hyperparameters of the LSSVM, constructing a high-performance classifier. Results Experimental results show that the optimized LSSVM model achieves an accuracy of 80.91% in underwater target recognition, which represents a 12.66% improvement over the non-optimized model. While achieving comparable accuracy, the proposed method significantly reduces optimization time by 95.62% and 95.10%, respectively, compared to randomized population-based optimization and deep neural network methods.Conclusions The multi-dimensional feature fusion and target recognition method demonstrates remarkable effectiveness, substantially enhancing the performance of underwater sonar image target identification and showing important practical application value.