In this paper, we propose a deep learning model based on convolutional neural network and recurrent neural network, which uses genome sequence data to identify human circular RNA splicing sites. Firstly, we preprocessed the original genome sequences and designed 16 models with two network depths, eight convolution kernel sizes and three LSTM parameters; secondly, the pooling layer was further tested for average pooling and maximum pooling; and GC content was added to improve the prediction ability of the model; finally, we predicted the circRNA in human seminal plasma. The results show that the model with convolution kernel of 32 × 4, depth of 1 and LSTM parameter of 32 has the highest recognition rate of 0.9824 on training data set, and 0.95 on test data set. Also, we tested our model with a published study and the accuracy reaches 0.83. The model has good performance in the recognition of human circular RNA splicing sites.