The pose estimation of underwater targets is highly valuable for the autonomous operation of underwater robots, environmental perception and docking of underwater equipment. To address the sparse and dispersive edge features of underwater target point clouds, a three-dimensional registration model that integrates multi-scale point cloud features is proposed. First, K-nearest neighbours maps of different scales in the point cloud are calculated, and point cloud features of different scales are extracted using the feature extraction module. Then, the extracted features of different scales are fused into the global features, and new features are generated by gradual fusion. Finally, the new features after fusion are fed into the pose estimation module to calculate the relative pose. In the underwater environment, a binocular camera and a case segmentation method are used to locate the target and obtain the target point cloud. The relative pose of the target point cloud obtained by segmentation is calculated. The experimental results show that the proposed method is obviously superior to the ICP and PCRNet methods.