A Neural Network Algorithm of Learning Rate Adaptive Optimization and Its Application in Emitter Recognition

The setting of the learning rate in neural network training is very important. A too low learning rate will reduce the network optimization speed and prolong the training time while a too high learning rate is easy to exceed the optimal value, leading to the difficulty of model convergence. To solve this problem, based on the analysis of two common learning rate strategies, the attenuating learning rate and the adaptive learning rate, combined with the Adam algorithm, this paper proposes an adaptive learning rate algorithm based on the value of the current loss function and the previous one, and verifies the effectiveness of the algorithm by using the actual radiation source signal. The experimental results show that compared with the Adam algorithm, the number of network training iterations is reduced by 45.5% and the recognition accuracy has increased by 3.6%, which effectively improves the learning speed and reduces the training time.