Addressing the issue of low accuracy in single tree segmentation of forest point clouds in complex spatial environments, this study focuses on two sample areas, namely coniferous and broad-leaved forests in Haikou Forest Farm, Xishan District, Kunming City. Utilizing airborne LiDAR point cloud data, a seed-based optimization algorithm is proposed for single tree segmentation. The method combines farthest point sampling and K-nearest neighbor search to identify tree top points, optimizes seed point selection through K-means clustering, and utilizes the Expectation-Maximization algorithm to determine the optimal parameters of a Gaussian Mixture Model, thereby improving its fitting accuracy. Finally, the single tree point cloud segmentation results are obtained based on the optimized Gaussian Mixture Model. Experimental results demonstrate the effectiveness of the proposed method, achieving overall accuracies O_accu(%) of 89.98% and 90.43% for the two study sites, respectively. The method exhibits good performance in accurately segmenting individual trees, distinguishing between connected tree crowns and pseudo-tree top points, and achieving precise segmentation of forest tree point cloud data.