A Neural Network Feature Enhancement Method Based on Feedback Compensation Mechanism

In traditional convolutional neural networks, the calculation process of input information is generally regarded as the process of feature extraction and representation. The effects of the models are closely related to the number of extracted features. In the research of this paper, the feature extraction process of neural network is regarded as a signal processing process. By using the feedback compensation mechanism of weak signal detection in the signal system, the output at the current time is fed back to the current input for information compensation, so as to achieve the effect of feature enhancement. This method is tested on MINIST data set and the experimental results show that the neural network with feedback compensation, without adding more parameters, can effectively improve the convergence speed of the model, reduce the fluctuation of loss function, and improve the accuracy. The comparison results show that the neural network with feedback compensation mechanism achieves the effect of feature enhancement.