Intelligent Acquisition and Processing of Weak Signals in High Noise Environments Based on Deep Neural Networks

In this paper, according to the characteristics of weak signals, A/D conversion and software programming are used to produce an intelligent acquisition device for weak signals. After completing the acquisition of weak signals, they are saved in the form of data sets and divided into training and testing sets, and the long and short-term memory network and convolutional self-encoder are used to construct a weak signal processing model and design the corresponding loss function. Simulation is used to confirm the effectiveness of the model in this paper. It is found that when the input voltage is 0.4V, it leads to the largest conversion error value, whose value is 0.00019357003601074. The model was tested, and it was concluded that the final loss values of the training set and the test set were reduced to 0.000112 and 0.00298, respectively. The model in this paper improves the signal with noise from -5dB to 16.04dB, and it perfectly removes the noisy information in it. The noise reduction ability of the CAE-LSTM network is much better than other control models. This study is capable of rejecting interference signals in weak signals with perfect accuracy, which is useful for academic research in the field of signals.