Abstract: The detection and protection of subsea pipelines and cables are crucial for ensuring energy security, maintaining communication reliability, and promoting ecological conservation. However, single-detection methods often face limitations in the complex seabed environment, resulting in restricted coverage and insufficient accuracy. To address this, this study proposes a method that fuses multibeam sonar point cloud data with side-scan sonar backscattered intensity images to improve the accuracy and robustness of pipeline and cable detection. The process primarily involves three steps: First, the contrast of side-scan sonar images is enhanced using contrast limited adaptive histogram equalization (CLAHE) to improve the data quality. Second, the feature points from both types of sonar images are extracted using the scale-invariant feature transform (SIFT) algorithm. High-precision image alignment and fusion are then achieved by combining the brute-force matcher (BFMatcher) with the random sample consensus (RANSAC) algorithm. Finally, based on the fused images, the YOLOv7 deep learning model is used for automatic identification and extraction of subsea pipelines and cables. This method, through its optimized network structure and enhanced dataset, enables effective fusion and recognition of multi-scale features. Experimental results reveal that the fused image significantly improves the identification precision of pipelines and cables. Furthermore, the detection confidence shows a peak increase exceeding 22.7% compared to using a single approach, and the accuracy attains 0.92. In addition, the method effectively mitigates both missed and false detections. This study not only provides technical support for the accurate detection of exposed subsea pipelines and cables but also offers an innovative approach for multi-source data fusion applications in complex seabed environments.