Research On Weak Signal Detection Algorithm Under A Small Number Of Samples

With the rapid development of science and technology,wireless communication has been applied in various fields.People have more and more needs for communication and higher requirements for signals.With the development of communication field,some signals which could not be studied due to the limitation of previous technology come into people’s field of vision,such as weak signals submerged in background noise.The study of weak signals has broadened people’s application range of signals.With the rise of neural network,researchers also begin to apply it in the field of weak signal detection,and has made remarkable progress and effect in the field.Although neural network has strong robustness and adaptive ability in the field of weak signal detection,it is limited by the number of training set samples.When the number of samples that can be provided for neural network learning is insufficient,the weak signal detection method with good performance due to neural network will become invalid.Therefore,it is necessary to study the weak signal detection method under the condition of a small number of samples.In view of the above problems,this paper first studies the weak signal detection method based on wavelet denoising data enhancement,and uses wavelet transform to remove the noise component in the sample signal,which can obtain a new signal that retains the original signal characteristics,and then expands the training sample dataset by selecting different wavelet basis functions,different wavelet decomposition layers and different wavelet threshold functions,and uses the weak signal detection model based on wavelet-long short-term memory network to simulate and verify the dataset expansion effect.The simulation results show that the weak signal detection method enhanced by wavelet denoising data has a good effect in data set expansion,and also has a good detection accuracy in weak signal detection.Secondly,the weak signal detection method based on Alex Net network transfer learning is studied,which uses time-frequency transformation to convert weak signals into two-dimensional images,and then uses Image Net data to pre-train Alex Net networks to obtain transferable network parameters,so as to realize the training and adjustment of weak signal detection model parameters under a small number of image samples.Combined with the idea of wavelet denoising,an improved weak signal detection method based on double wavelet-Alex Net network is proposed,and its detection performance is simulated and verified.The simulation results show that the improved weak signal detection method based on double wavelet-Alex Net network has good detection accuracy under the condition of a small number of samples,and can maintain more than 80% detection accuracy when the signal-to-noise ratio is-20 d B.