Millimeter wave fuze obtains the detailed structure information by transmitting a wideband signal. However, in the process of non-cooperative detection and recognition, a large number of target samples could not be fully utilized due to the lack of target class information. To this end, Laplacian Score(LS) is extended to Label Reconstruction based Laplacian Score(LRLS) and applied to the case of semi-supervised learning. Under the framework of LS, LRLE utilizes the label reconstruction technique to calculate the Laplacian matrix. In order to better describe the similarity between the high-dimensional samples, the Euclidean distance is replaced with the geodesic distance. The experimental results show that LRLS performs better than the traditional methods.