Multi-target Ground Object Recognition Based On Deep Learning

Lidar is a radar system which actively emits a laser beam to detect the position,direction and attitude of the target and receives information through the echoes reflected from the target.Because the data it receives has high density,large data and scattered features,it is called Point cloud data.Point cloud data can be used for urban scene reconstruction,driverless,environmental testing and more.Due to the unstructured and disordered nature of point cloud data,most current methods based on lidar point cloud recognition use manual extraction features,which consume a lot of cost and are difficult to extract high-level information.In order to solve the above problems,this paper proposes a deep recognition based lidar multi-target ground object intelligent recognition method based on deep learning.Firstly,the point cloud data collected by the UAV lidar,and then uses the point cloud spatial position information to divide the entire point cloud scene into regions,divides the point cloud scene into a single element,and sets it.The voxel number threshold parameter and the distribution of various types of points in the statistical voxel complete the construction of the data set and the label of the corresponding voxel.Then a method of cascading classification is proposed in this paper.The feature learning,cross-validation,and recognition of each training set are carried out through the cascade of threedimensional convolutional neural network model and neighborhood algorithm model,and the recognition results are analyzed in detail..Finally,the experiment tests and optimizes the cascade model using outdoor point clouds of different locations and different densities.The experimental results show that the proposed method has a good performance on accuracy and efficiency.