DIM point cloud filtering is a key step in feature classification,feature monolithic extraction and terrain feature analysis.In order to solve the difficulty and low accuracy of DIM point filtering due to complex terrain in highland mountains,an improved PTD filtering algorithm is proposed.The mountainous areas on the southern edge of the dinosaur valley in the Central Yunnan Plateau are chosen as the experimental area.Firstly,DJI Phantom 4 RTK is used to collect the image data and solve the dense image to obtain the DIM point cloud.Then,considering the large difference in elevation between the summit and the foot of the mountain,and the ground points on both sides of the mountain are easily identified as non groundpoints,the topographic feature point of ridge valley intersection is the seed point of PTD filtering algorithm,and the improved PTD filtering algorithm refine the irregular triangular network on both sides of the mountain.The results show that:1) the ground points obtained by the PTD filtering algorithm are more completely preserved throughout the experimental area,but obvious features such as low vegetation on both sides of the mountain and vegetable trellises at the foot of the mountain are basically not eliminated,and the ground points on parts of the mountain are easily identified as non ground points and hollow phenomena in the K1,K2 and K3 areas of the mountain.2) For the complex terrain of the mountainous area on the southern edge of the dinosaur valley,the ridge valley intersection terrain feature points are proposed as the seed points of the PTD filtering algorithm.On both sides of the mountain refinement network,the mountain low vegetation partly cleared.Vegetable shed large area is cleared compared to the PTD filtering algorithm.The ground points on both sides of the mountain are preserved,and no obvious point cloud voids are observed.