Abstract: Feature selection and classifier design are crucial steps in radar emitter signal sorting. This paper proposes a signal sorting algorithm based on Sparse Auto Encoder (SAE) feature optimization and Bagging ensemble learning. First, 16-dimensional features are extracted from intra-pulse, inter-pulse, and time-frequency transform domains of radar pulses to form a feature vector. This vector serves as input to establish an SAE-based feature optimization model, which automatically selects a 5-dimensional optimal feature subset to eliminate redundant information and enhance computational efficiency for subsequent sorting algorithms. Subsequently, the optimal feature subset obtained by SAE is utilized as input to construct a Bagging ensemble learning model. This model employs K-means, Mean-shift, and Gaussian Mixture Model (GMM) clustering methods as base learners and DBSCAN as a meta-learner for sorting identification, ultimately generating final sorting results. Simulation experiments validate the sorting performance of the proposed method. Results demonstrate that compared to directly using high-dimensional raw features, the optimal feature subset selected by SAE significantly improves sorting performance. Additionally, the Bagging ensemble model achieves higher sorting accuracy, lower false alarm rates, and lower missing alarm rates compared to individual models.