提出一种基于Kohonen神经网络无监督的深度图像分割方法。首先计算了深度图像各点的导数,进而得到各点的法向,以法向和深度值作为每一点的特征矢量,引入自组织神经网络进行初始的聚类;为消除初始聚类产生的过分割现象,采取相邻表面片法向分析的方法进行再分割,得到最终的分割结果。本方法避免了通常的区域分割方法初始种子不易选取的弱点,聚类所用样本少,速度快。实验结果表现了算法良好的性能 An unsupervised depth image segmentation method based on Kohonen neural network is proposed. Firstly, the derivative of each point of the depth image is calculated, and then the normals of each point are obtained. The normal and depth values ​​are taken as the eigenvectors of each point, and the initial clustering is introduced by using the self-organizing neural network. Segmentation phenomenon, to take the adjacent surface normal method of subdivision to obtain the final segmentation results. The method avoids the weak points that the usual initial segmentation method of the region segmentation method is not easy to select, the sample used in the clustering is less and the speed is faster. The experimental results show the good performance of the algorithm