Dynamic target feature selection in pixel change space for array GM-APD lidar

For detecting dynamic targets, the relative position change between the target and array GM-APD lidar significantly affects the stability and the classification ability of the extracted features, thereby affecting the target classification and recognition. In this paper, a feature selection algorithm is proposed that maps features to the space of the calculated average number of pixel changes to enhance the classification ability of features. Divide the feature data according to different scales in this space to keep the target feature values close in each window, thereby reducing the feature instability caused by motion. At the same time, establish the loss function in each window considering the importance and relevance of features and the number of selected features. Use particle swarm optimization to adaptively select the optimal feature subset and determine the optimal window scale. The experimental results show that the proposed mean of peak ratio of the decomposition waveform feature shows significant classification ability and stability in dynamic target classification. The classification accuracy is significantly improved when dealing with different dynamic vehicle scenes using the feature subset selected by the proposed algorithm. Compared with the undivided window, the classification accuracy is improved by 9.1 %–53.1 %, and compared with the division time window, the classification accuracy is improved by 5.7 %–18.8 %. This study provides robust technical support for dynamic target classification and recognition.