Previous researches of fingerprinting feature extraction of frequencyhopping signals have just classified signals through deep learning. We wish to create a new model based on CNN convolutional neural network which can extract the characteristics of the pre-processed frequency hopping signals and classify them using those characteristics. Firstly, we perform short-time Fourier transform on the collected frequencyhopping signals to present them in frequency hopping sensitive frequency domain. The converted signals will be putted into the CNN network model and convolved, pooled and fully connected so that we can get final classification results. In this process, we use multi-layer convolution to extract deeplevel features in the frequency domain of the signals and apply Batch Normalization and Callback functions which can not only optimize and accelerate network convergence speed, but also prevent overfitting effectively. From the final data, the new network model has higher individual recognition accuracy rate than the previous.