With the continuous development and application of USV technology, its threat to ships is increasing. In order to improve the recognition performance of multi-component laser/infrared/millimeter wave detector on small surface targets, a composite detection signal recognition method MCCNN-XGB based on multi-channel convolutional neural network (Multi-Channel Convolutional Neural Network, MCCNN) and extreme gradient lifting decision tree (Extreme Gradient Boosting, XGBoost) is proposed. At the same time, a single channel CNN recognition network and XGBoost recognition algorithm based on artificial feature extraction are constructed as a comparison. Then, the target recognition performance of the above three models is evaluated and compared through the test data of UAV mount USV target. The test results show that the recognition algorithm based on MCCNN-XGB performs the best, with a test accuracy of 97.26%. The recognition method proposed in this paper can effectively extract the features of the complex detection signal, and can reduce the false recognition rate and missing recognition rate, which has a good recognition effect.