Research On Target Inversion And Classification Method Of Millimeter-Wave Synthetic Aperture Radiometer

The millimeter-wave synthetic aperture radiometer is a safe and efficient passive imager.Its working wavelength is between microwave and infrared,and it can penetrate media such as smoke,dust,plastic and clothing to achieve high-resolution imaging.It has broad application prospects in short-range non-destructive detection such as hidden object detection and guided navigation.However,due to the influence of near-field spherical waves,it is difficult for millimeter wave aperture synthesis radiometer to achieve high-precision near-range imaging inversion,and the resulting millimeter wave image quality is poor,which is not conducive to subsequent tasks such as target classification.These problems limit the practical application of millimeter wave aperture synthesis radiometer.Therefore,this paper focuses on the target inversion and classification method of the near-field millimeter-wave synthetic aperture radiometer.The main research content of this paper is as follows:(1)Due to the large number of interference factors in the near-field condition and the large error in the observation data,the near-field imaging effect of the millimeter-wave synthetic aperture radiometer is not ideal.To solve this problem,this paper combines the idea of deep learning with the principle of millimeter-wave imaging,and proposes a near-field imaging method based on full connection.This method utilizes the full connectivity between the visibility function and the brightness temperature distribution of the target scene,uses a fully connected feed-forward neural network,inverts the millimeter wave image from the sparse visibility function,and then combines the dense residual network to complete the denoising of the millimeter wave image and enhanced.The experimental results show that,compared with the MFFT algorithm and the G-matrix algorithm,the root mean square error and peak signal-to-noise ratio are increased by 28.2%,11.4%,and 29.9%,16.7% respectively under the conditions of no noise and noise;the imaging inversion time shortened by 88.8%.The millimeter-wave image inverted by this method has great advantages in imaging accuracy,noise suppression and reconstruction time.(2)Aiming at the problem that the synthetic aperture radiometer is greatly disturbed by noise in the near-field imaging process,information is lost in the inversion imaging process,and the classification accuracy of the traditional millimeter-wave target classification method based on the image domain is not ideal,this paper proposes a visibility-based The target classification method of the function uses the neural network algorithm to establish the mapping relationship between the visibility function and the target label,which can not only avoid the systematic error and noise interference introduced by the imaging inversion,but also make full use of the phase and amplitude information in the original visibility function,effectively improving the target classification accuracy of the millimeter-wave synthetic aperture radiometer.The experimental results show that under the three noise conditions,the classification accuracy of the standard dataset and the simulated dataset reaches 98.41%,92.26%,88.72% and 93.71%,90.13%,86.74%,respectively.Compared with the traditional target classification algorithm in mm Wave image domain,the classification accuracy and anti-noise performance of this method are significantly improved.