LiDAR point cloud data acquisition is often affected by reflective objects such as glass and mirrors, which produce specular reflections and introduce mirror reflection noise. Due to the similarities between mirror reflection noise and the corresponding physical point clouds, traditional denoising methods struggle to effectively remove such noise points. To address this challenge, this paper proposes a method for identifying and removing mirror reflection noise based on the similarity and difference between the two. First, the scene point cloud data is divided into several point cloud blocks, and the mirror symmetry plane is extracted through the random sampling consensus algorithm, and then the Euclidean distance from the point cloud block to the mirror symmetry plane is judged to preliminarily screen out the mirror reflection noise, and the 3D point cloud registration is performed The similarity detection with the two-dimensional depth image accurately identifies the mirror reflection noise, and finally the point cloud convex hull is used to calculate the density value combined with Euclidean clustering to remove mirror reflection noise. The experimental results show that this paper can accurately identify the mirror reflection noise and remove it effectively.