Aiming at the issues of low resolution, significant noise interference in sonar images, and insufficient accuracy of single data sources in submerged pile structure defect detection, an improved Murphy evidence theory-based multi-source sonar image fusion detection method is proposed. Firstly, a complementary model based on Darknet-53 and ResNet-50 is constructed to provide diversified evidence and strengthen information representation. Secondly, the squeeze-excitation(SE) attention mechanism is integrated into the two complementary models to enhance the detection performance for defect areas. Finally, the Murphy evidence theory is modified to facilitate data fusion, thereby improving the accuracy and robustness of the method. Experimental results show that the proposed method outperforms other models, with precision, recall, and average precision all exceeding 93.7% for detecting common defect types. These results demonstrate that the improved Murphy evidence theory, integrated with deep learning and multi-source data, significantly improves the accuracy and environmental adaptability of sonar image defect detection, offering an innovative solution for defect identification in complex underwater engineering projects.