Abstract: Since unmanned aerial vehicle (UAV) and bird both belong to the category of "low, slow, and small" (LSS) targets, accurate identification and classification of them are particularly important. However, current commonly used classification methods suffer from a "black box" problem, where the decision basis of the model cannot be clearly interpreted. To address this issue, starting from the motion principles of UAV and bird, the differences in their motion trajectories are analyzed, the corresponding motion features are extracted, and these features are further categorized into macroscopic and microscopic characteristics. Subsequently, a light gradient boosting machine algorithm is employed to construct a classification model. The tree-structured Parzen estimator is introduced to optimize model parameters, while the Shapley additive explanations interpretation algorithm is integrated to visually demonstrate the model′s decision-making process, thereby clarifying the contribution of different motion features to model accuracy and highlighting the importance of each feature in classification decisions. Experiments conducted on the "passive radar LSS target detection dataset" show that the proposed UAV and bird recognition model achieves an accuracy rate of 92 %. Comparisons with other algorithms under identical conditions confirm the effectiveness and superiority of the proposed method.