ISAR (inverse synthetic aperture radar) imaging method based on convolutional neural network

Abstract

The invention discloses an ISAR (inverse synthetic aperture radar) imaging method based on a convolutional neural network, and the imaging method comprises the following steps: S1, constructing an ISAR data set; S2, obtaining an initial image by two-dimensional Fourier transform; S3, constructing the convolutional neural network; S4, constructing a training set and a verification set of the convolutional neural network; S5, updating the parameters of the convolutional neural network through using a supervised learning method; S6, using the trained convolutional neural network to achieve the ISAR downsampling data imaging. The convolutional neural network of the invention can be used for extracting more feature information and can effectively avoid the gradient dispersion phenomenon, thereby reconstructing a higher quality ISAR image.

Description

technical field [0001] The invention belongs to the technical field of radar signal processing, and relates to an inverse synthetic aperture radar imaging method based on a convolutional neural network. Background technique [0002] Inverse synthetic aperture radar is a typical imaging radar system, which is mainly used to obtain high-resolution images of non-cooperative moving targets, and is an effective means of target recognition. The traditional radar imaging method is the range-Doppler (RangeDoppler, RD) type imaging method, which uses pulse compression technology to obtain high resolution in the range direction, and uses coherent accumulation time (ie coherent processing time, Coherent Processing Interval, CPI) The Doppler modulated echo signal obtains high resolution in azimuth. However, the RD imaging method usually needs to make some assumptions about the echo signal, imaging scene and imaging configuration, such as small aperture, small scene, etc. When the assum...

AI Technical Summary

Problems solved by technology

However, the CS inverse SAR imaging method still faces the following problems: 1. Sparse representation; for inverse SAR imaging, some preset image transformations (such as wavelet transform) are generally used to find the sparse representation. The representation is not optimal, so it limits the quality of the CS imaging method

2. Random measurement matrix construction; the measurement matrix must satisfy the constraint equidistant condition, or be irrelevant to the sparse basis. The determination of the measurement matrix affects the required sampling data and whether the target scene can be accurately reconstructed in the end.

Third, the reconstruction method is too inefficient to be practical; the CS reconstruction algorithm is basically a series of convex optimization methods or iterative greedy algorithms proposed for the minimization of the L1 norm

This iterative optimization imaging algorithm has low computational efficiency, requires more storage resources, and is difficult to meet real-time processing requirements.

Images