The invention discloses a radar radiation source identification method based on bispectral self-coding features. The method comprises the following steps that step 1, radar radiation source signals are acquired, and preprocessing and noise filtering are performed on the radar radiation source signals; step 2, the direct estimation method is applied to the discrete radiation source signals obtainedby sampling to calculate the bispectral image of the discrete signals; and step 3, feature learning is performed on the extracted bispectral image by using an over-limit learning machine sparse self-coding algorithm, and finally a radar radiation source identification model is constructed by using an over-limit learning machine classification algorithm. The integrated radar radiation source classification framework based on bispectral feature learning and over-limit learning machine identification is constructed so that the rapid and efficient radar radiation source signal identification method can be established.