Abstract: Sea state classification is the basis of ocean monitoring and maritime target detection. Sea state monitoring technology based on radar echo has significant advantages of all-day and all-weather. In view of the problem that the current radar sea state classification cannot balance between accuracy and timeliness, a sea state classification method based on radar high resolution range profile(HRRP) fractal features and residual convolutional cyclic sequence classification network (ReSeqNet) is proposed in this paper. Firstly, the q-order Hurst exponent and singularity power spectrum of the measured sea clutter HRRP sequence are extracted based on fractal theory, and the fusion features are constructed. Secondly, the ReSeqNet network based on the convolutional neural network-recurvent neural network architecture is designed to capture the local fractal features and their time correlation to realize sea state classification. Finally, the performance of the proposed network is tested using a measured data set. The experimental results show that the proposed method can achieve three-level classification of low sea states (2-level sea states and 3-level sea states), 4-level sea states and 5-level sea states. The accuracy of the two-level classification of high and low sea states is 100%, and the classification accuracy of high sea states (4-level sea states and 5-level sea states) is 99.4%, which is about 6% higher than the existing method under the condition of shorter echo accumulation time.