A novel spectral imaging method based on multi-modes in time-frequency domain is proposed. This method is used for the nondestructive detection of four kinds of ceramic matrix composite (CMC) samples with defects. A THz image set for the test samples is formed, and five quantity evaluation indexes for image quality are introduced. The THz images with the best quality are chosen via the index integration treatment. The retrieval of this image is realized based on the Scale Invariant Feature Transform (SIFT) feature selection and K-means clustering algorithm. The experimental results confirm that the proposed method can be effectively used in the imaging detection of the CMC materials with various defect widths at different positions.