New Problems such as redundancy bispectrums, cross-items and possess too high feature dimension still arise though the existing methodes of bispectrum feature extraction can meet the needs of signal classification and recognition at present. A new bispectrum cascade feature extraction method is proposed. Firstly, the bispectrum is converted to gray image, and the amplitude of bispectrum is denoted by gray value. Secondly, the bispectrum cascade feature is extracted based on image processing technique. The gray level co-occurrence matrix (GLCM) cascade feature can specify the bispectrum image texture information of radar emitter signal. Finally, the comparative experiment is made. The result shows that the GLCM features are superior to Hu-invariant moment features greatly in the aspect of performing radar signal classification and recognition.