Sea Target Detection On Deep Learning

The existence of sea clutter is the reason why the detection of small targets on the sea has always been a matter of concern and difficulty in radar signal processing.The echoes of the sea target are mainly from the sea surface's own clutter and the target's scattered echo.The difficulty with direct detection of small targets is that the sea clutter background will annihilate the echo signal of the small target.The indirect detection of sea clutter has certain research value for the detection of small targets.In this paper,the target detection in sea clutter is studied from the perspective of one-dimensional signal processing and two-dimensional image processing,respectively.The main work is as follows:1.One-dimensional sea clutter sequence prediction based on stack self-encoder are investigated.By analyzing and studying the properties of IPIX real sea clutter,the short-term predictability of one-dimensional sea clutter sequence can be proved.Based on the short-term predictability,a stack self-encoder prediction model is established.Simulation results show that the model is effective and the prediction accuracy is improved compared to other methods.2.Target Detection Based on Convolutional Self-Encoder in 2D Image Domain is studied.For the time-distance image data set of simulated sea clutter echo data,feature extraction and target detection are performed using a convolutional auto-encoder.SIRP method is used here to simulate compound K-distribution sea clutter under different sea condition.Test results show that the accuracy of the proposed method improves detection accuracy.3.Removal of noise in sea clutter image based on sparse denoise self-encoder is examined.A sparse noise denoising autoencoder is designed to remove noise from noise-contaminated sea clutter images.Simulation results prove that the proposed method shows good denoising performance,which is verified by several indicators.