CN107085829B - Frequency spectrum associated super-resolution method for broadband electromagnetic distribution detection - Google Patents

Abstract

A spectral correlation super-resolution method for broadband electromagnetic distribution detection, which can realize the restoration of images formed by broadband radiation sources of 1-6 GHz. Firstly, a model is established in the simulation software according to the degraded image of the unknown radiation source, and the point spread functions of different frequencies are obtained, and then the two-dimensional Fourier transform of the space is performed on the point spread functions of different frequencies, and the cut-off frequency width in space is recorded. Then draw the corresponding frequency-width curve; perform two-dimensional Fourier transform on the acquired broadband electromagnetic distribution image and obtain the cut-off frequency width in space, and determine the highest frequency of the radiation source on the curve obtained by simulation; The point spread function corresponding to the highest frequency component is used to process the image, and the L_R iteration method in the narrowband super-resolution algorithm is used to restore the degraded image. The invention overcomes the incompatibility between the traditional narrow-band super-resolution algorithm and the electromagnetic distribution detection system, can identify the highest frequency in the broadband radiation source and effectively improves the resolution of the image.

Description

A Spectral Correlation Super-Resolution Method for Broadband Electromagnetic Distribution Detection

technical field

The invention relates to a broadband super-resolution method for a broadband electromagnetic distribution detection system, in particular to the field of broadband electromagnetic detection and the field of image processing.

Background technique

At present, the detection method of electromagnetic distribution mainly adopts the form of antenna and receiver to perform point-by-point manual detection. This method has many disadvantages, such as the detection speed is slow, and the antenna will interfere with the original field distribution. The detection of electromagnetic distributions using parabolic reflection surfaces can be achieved quickly and without interference. Affected by the size of the antenna, the diffraction of the broadband electromagnetic distribution detection system is severely limited, and how to improve its resolution so that the electromagnetic distribution can be clearly distinguished has become a new difficulty.

Due to the limited diffraction of the broadband electromagnetic distribution detection system, the imaging is blurred, and super-resolution methods are needed to improve the resolution. In recent years, the super-resolution algorithm can be divided into single-frame super-resolution algorithm and multi-frame super-resolution algorithm based on the number of imaging frames, and can be divided into frequency domain algorithm and spatial domain algorithm based on the operation method, and the spatial domain algorithm can be subdivided into non-uniform difference. method, statistical theory method, convex set projection method, regularization method, blind reconstruction method, etc. The core of the super-resolution algorithm with better recovery effect is to determine a more accurate point spread function, so these super-resolution algorithms can only process the image formed by a single frequency radiation source, while the radiation formed by the broadband radiation source is in the image. The frequency components of the source are complex, and the point spread function cannot be determined, so the existing algorithms are difficult to meet the requirements.

SUMMARY OF THE INVENTION

The technology of the present invention solves the problem: overcomes the deficiencies of the prior art, and provides a method that the present invention cannot effectively solve the problem of image blurring in the broadband electromagnetic distribution detection system for the existing narrow-band super-resolution algorithm. Based on the existing correlation, the high-frequency components are used to restore the image, and the low-frequency components are located to restore the blurred image formed by the broadband radiation source. A spectral correlation super-resolution algorithm for broadband electromagnetic distribution detection is proposed.

In simulation and actual tests, it is found that if there are two or more radiation sources with different frequencies at the same location, or there are broadband radiation sources, the point spread function corresponding to the highest frequency in the radiation source is used to adopt the narrowband ultra The resolution algorithm to restore the image can achieve better results. Therefore, the key in the spectral correlation algorithm is how to obtain the specific value of the high frequency frequency at this position, so as to obtain the corresponding point spread function for spectral correlation super-resolution processing.

The spatial characteristics of the point spread function corresponding to different frequencies are different. Figure 2 and Figure 3 are the point spread functions corresponding to the 1GHz frequency and the 3GHz frequency obtained by simulation, respectively. From the perspective of the spatial domain, the higher the frequency, the narrower the lobe width of the point spread function, and such differences will also form corresponding features in the spatial frequency domain, and the determination of the high-frequency components in the spectral correlation super-resolution algorithm is exactly based on such characteristics.

The space-frequency domain analysis is performed on the spread functions of two different frequency points, as shown in Figure 4 and Figure 5. From the figure we can see that the higher the frequency point spread function the higher the spatial cutoff frequency. The number and location of radiation sources does not affect the cutoff frequency in the airspace. In the spatial domain, if there are different frequency components, the spatial spectrum after Fourier transform is also the direct superposition of the spatial spectrum of each frequency component. Therefore, the highest frequency in an image can be determined by the high-frequency cut-off frequency, that is, the image to be processed is subjected to two-dimensional Fourier transform, and its high-frequency cut-off frequency is obtained in the space frequency domain. Process the exact size of the highest frequency components in the image. The image is then restored using a narrow-band super-resolution algorithm.

The spectrum correlation super-resolution algorithm used in the broadband electromagnetic distribution detection of the invention can realize the restoration of the image formed by the broadband radiation source of 1-6 GHz, and overcome the incompatibility of the traditional narrow-band super-resolution algorithm with the electromagnetic distribution detection system. Specifically, the method of the present invention is implemented through the following steps 1 to 3.

Step 1, establish a simulation model in the simulation software, obtain the point spread functions of different frequencies, perform a two-dimensional Fourier transform of the space on the point spread functions of different frequencies, record the cut-off frequency width in space, and then draw the corresponding Frequency-width graph;

Step 2: Obtain a broadband electromagnetic distribution image through a parabolic reflector and an electric field probe photoelectric conversion system, perform a two-dimensional Fourier transform on the broadband electromagnetic distribution image obtained by the image, and obtain the cut-off frequency width in space, which is obtained by simulation in step 1. Determine the highest frequency of the radiation source on the frequency-width curve of ;

Step 3: Use the point spread function corresponding to the highest frequency component to process the image, and use the L_R iteration method in the narrowband super-resolution algorithm to restore the degraded image.

In the step 3, the fast L_R iteration method in the narrowband super-resolution algorithm is used to perform image restoration on the degraded image, and the specific steps are as follows:

(1) Input the degraded image I to be restored, input the point spread function PSF corresponding to the component of the highest frequency, input the number of iterations NUMIT, and initialize the restored image;

(2) Initialize the restored image J{1}=I, if the sampling rate of the point spread function PSF is higher than the degraded image I, then the point spread function is further sampled, so that the sampling rate is the same as the input image I;

(3) Using the maximum likelihood method, the iterative kernel is determined as:

where J ^t+1 is the restored image obtained by the latest iteration, J ^t is the image obtained by the previous iteration, PSF is the point spread function, and B=Possion(J*PSF) is the Poisson process of image degradation;

(4) Use the iterative kernel in step (3) to iterate the input image, and perform regularity check on each result;

(5) Stop the iteration according to the number of input iterations, and output the restored image J obtained by the latest iteration.

The advantages and positive effects of the invention are: the invention overcomes the incompatibility between the traditional narrowband super-resolution algorithm and the electromagnetic distribution detection system, can identify the highest frequency in the broadband radiation source and effectively improves the resolution of the image. The traditional narrow-band super-resolution algorithm cannot be directly applied to the image processing of broadband electromagnetic detection. The spectral correlation super-resolution algorithm proposed by the present invention uses the cut-off frequency width in the spatial frequency domain to determine the highest frequency component, and then determines the high frequency point spread function. , the simulation shows that the cut-off frequency width and the radiation source frequency have a good linear relationship; after the point spread function of the highest frequency is determined, the traditional L_R iterative method is used to restore the blurred image, and the high frequency component is used to locate the low frequency. The problem of blurred image formed by broadband electromagnetic distribution detection system. Simulation and experimental results show that the method of the present invention can effectively improve the resolution of the image when there are multiple radiation sources with different frequencies or broadband radiation sources at the same position.

Description of drawings

Fig. 1 is the overall flow chart of the spectral correlation super-resolution method for broadband electromagnetic distribution detection of the present invention;

Fig. 2 is the point spread function diagram of 1GHz obtained by simulation;

Fig. 3 is the point spread function diagram of 3GHz obtained by simulation;

Fig. 4 is the space spectrogram after the point spread function of 1GHz undergoes two-dimensional Fourier transform;

Fig. 5 is the space spectrogram of the point spread function of 3GHz after the two-dimensional Fourier transform;

6 is the degraded electromagnetic distribution image obtained by the simulation example of the present invention;

Fig. 7 is the spatial spectrum diagram of the degraded electromagnetic distribution image after two-dimensional Fourier transform;

Fig. 8 is the corresponding diagram of radiation source frequency and cut-off frequency width obtained by the feed-forward parabolic reflector;

Fig. 9 is the corresponding diagram of radiation source frequency and cut-off frequency width obtained by the biased parabolic reflector;

Figure 10 is an image processed by this method.

Detailed ways

The method will be further described in detail below in conjunction with the accompanying drawings and examples.

The invention proposes a spectrum correlation super-resolution algorithm for broadband electromagnetic distribution detection. The method of the invention realizes the improvement of the resolution of the blurred image formed by the broadband radiation source.

As shown in FIG. 1 , the spectrum correlation super-resolution algorithm for broadband electromagnetic distribution detection according to the present invention includes steps 1 to 3 .

Step 1, establish a simulation model in the simulation software, obtain point spread functions of different frequencies, and then perform two-dimensional Fourier transform of space on the point spread functions of different frequencies, record the cut-off frequency width in space, and then draw the corresponding The frequency-width curve of .

In the embodiment of the present invention, the electromagnetic distribution images of different frequencies are obtained in the simulation software feko, and certain criteria need to be established when selecting the cut-off frequency width. It can be found by simulation that the cutoff frequency width has a very good linear relationship with the frequency of the radiation source. The received image pixel matrix is subjected to a two-dimensional Fourier transform, and then its spectral lines are observed from two directions. Record the cutoff frequency width of the spatial frequency spectrum of the two directions corresponding to different radiation sources. Since the minimum value on the spectral line cannot be zero, in order to determine the cutoff frequency more accurately, the selection method of the cutoff frequency is specified: near the last minimum value point in the spectral line of the central section, the first position that appears less than 1 is Specified as the cutoff frequency point. The data obtained by recording the simulation is drawn into a curve, as shown in Figure 8. FIG. 8 is a corresponding diagram of the radiation source frequency and the cut-off frequency width obtained by the example of the present invention. Next, the paper performs a similar simulation calculation for the case of bias. The offset reflector also uses a parabolic reflector with a diameter of 3m and a focal length of 1.7m. A dipole is placed at the position of (0, 10, 0) m, and the frequency of the dipole is changed from 500MHz to 6GHz. The simulated frequency interval is 500MHz, and the image is received at the ideal image plane. The receiving surface is -1 to 1m in the X direction, and a sampling point is set at an interval of 0.01m; in the Z direction, a sampling point is set at a distance of -1 to 1m, and an interval of 0.01m, with a total of 201*201 pixels. The received degraded blurred image pixel matrix is subjected to two-dimensional Fourier transform, and then its spectral lines are observed from two directions respectively. Record the width of the cutoff frequency corresponding to the different radiation sources in the two directions. However, during the calculation, it was found that the spectral lines in the vertical direction were chaotic, and it was basically impossible to accurately determine the position corresponding to the cutoff frequency, while the position of the cutoff frequency in the horizontal direction was relatively clear. This phenomenon was caused by the biased parabolic reflector itself. caused by the asymmetry. The corresponding relationship between the cutoff frequency width and the radiation source frequency in the horizontal direction is shown in Figure 9. FIG. 9 is a corresponding diagram of the radiation source frequency and the cut-off frequency width obtained by the biased parabolic reflector.

It can be seen that there is a similar proportional relationship between the width of the cutoff frequency and the frequency of the radiation source in the case of bias. And when the actual focal-to-diameter ratios of the feed-forward and biased reflecting surfaces are the same, their cut-off frequency widths at the same radiation source frequency in the horizontal direction are also basically the same. Therefore, an appropriate focal-to-diameter ratio is selected for simulation to obtain a broadband electromagnetic distribution image.

In step 2, a two-dimensional Fourier transform is performed on the acquired broadband electromagnetic distribution image to obtain the cut-off frequency width in space, and the highest frequency component of the radiation source is determined on the curve obtained by the simulation.

Through simulation, it can be found that the spatial cutoff frequency width is related to the shape of the system reflector, the highest frequency and the polarization direction, but has nothing to do with the form and number of radiation sources. In this way, the present invention establishes the corresponding relationship between the estimated cut-off frequency width and the frequency of the radiation source, so that the highest frequency component of the radiation source can be identified to a certain extent, and then the point spread function can be determined for further processing.

However, the received image has a limited signal-to-noise ratio due to the limited sensitivity of the electric field sensor probe. The Lucy-Richardson iterative algorithm can effectively suppress the noise in a certain signal-to-noise ratio image, but the existence of system noise will affect the estimation method of the frequency and location of the radiation source mentioned in this chapter. Therefore, denoise the degraded image first, so that its spatial frequency spectrum becomes clean and the cutoff frequency judgment is easier.

Step 3: Use the point spread function corresponding to the highest frequency component to process the image, and use the L_R iteration method in the narrowband super-resolution algorithm to restore the degraded image.

The specific algorithm process is as follows:

Step 1: Input the degraded image I to be restored, input the point spread function PSF of the electromagnetic imaging system, input the number of iterations NUMIT, and initialize the restored image;

Step 2: Initialize the restored image J{1}=I, if the sampling rate of the point spread function PSF is higher than that of the degraded image I, then further subsampling the point spread function to make the sampling rate the same as the input image I;

Step 3: Determine the iterative kernel of the algorithm as:

where J ^t+1 is the restored image obtained by the latest iteration, J ^t is the image obtained by the previous iteration, PSF is the point spread function, and B=Possion(J*PSF) is the Poisson process of image degradation;

Step 4: Use the iterative kernel in step 3 to iterate on the input image, and perform regularity check on each result;

Step 5: Stop the iteration according to the number of input iterations, and output the restored image J obtained by the latest iteration.

A simulation example is used below to specifically illustrate the implementation of the method. The experimental system adopts a deflection-feed parabolic reflector with a diameter of 1.5 meters and a focal length of 1.1 meters. Two dipoles of different frequencies are placed adjacent to each other at a horizontal distance of 5 meters from the reflective surface. Assuming that the radiation frequency of the dipole is unknown, the image of the degraded electromagnetic distribution received on the focal plane after passing through the reflective surface is shown in Figure 6. The two-dimensional Fourier transform is used for the degraded image, and the spatial spectrum is obtained as shown in Figure 7.

The radiation sources of different frequencies are set in the simulation software, the degraded images are received at the focal plane, and the spatial frequency spectrum is obtained by two-dimensional Fourier transform, and their cut-off frequency widths are recorded. The obtained cut-off frequency width corresponds to the frequency The curve is shown in Figure 8, where the abscissa represents the frequency of the radiation source, and the ordinate represents the corresponding cutoff frequency width.

Through observation, it is found that the cut-off frequency width of the spatial frequency spectrum obtained by the two-dimensional Fourier transform of the degraded image is 40. On the curve, the frequency corresponding to the cutoff frequency width 40 is 3 GHz. The simulated image is the degraded image obtained by the 1GHz and 3GHz dipoles passing through the reflective surface. This also verifies that the method of determining the highest frequency component by spatial information is feasible. Using the point spread function of 3GHz, the L_R iteration method is used to restore the degraded image, and the obtained restored image is shown in Figure 9. Through the comparison between FIG. 6 and FIG. 9 , it can be clearly found that the resolution of the image has been effectively improved, which shows that the present invention achieves satisfactory results in estimating the highest frequency component and improving the image resolution.

The above embodiments are provided for the purpose of describing the present invention only, and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalent replacements and modifications made without departing from the spirit and principle of the present invention should be included within the scope of the present invention.

Claims (1)

1. A frequency spectrum associated super-resolution method for broadband electromagnetic distribution detection is characterized by comprising the following implementation steps:

step 1, establishing a simulation model in simulation software, acquiring point spread functions of different frequencies, performing spatial two-dimensional Fourier transform on the point spread functions of the different frequencies, recording cut-off frequency width in space, and then drawing a corresponding frequency-width curve graph;

step 2, obtaining a degraded broadband electromagnetic distribution image through a parabolic reflecting surface and an electric field probe photoelectric conversion system, performing two-dimensional Fourier transform on the degraded broadband electromagnetic distribution image obtained from the non-degraded image, and solving the cut-off frequency width in space, and determining the highest frequency of the radiation source on the frequency-width curve obtained by simulation in step 1;

step 3, processing the image by using a point spread function corresponding to the highest frequency component, and performing image recovery on the degraded broadband electromagnetic distribution image in the step 2 by using an L _ R iteration method in a narrow-band super-resolution algorithm;

in the step 3, the image recovery is performed on the degraded broadband electromagnetic distribution image by adopting a fast L _ R iteration method in the narrow-band super-resolution algorithm, and the specific steps are as follows:

(1) inputting a degraded broadband electromagnetic distribution image I to be restored, inputting a point spread function PSF corresponding to the component with the highest frequency, and inputting an iteration number NUMIT;

(2) making the initialized and restored image J {1} -, if the sampling rate of the point spread function PSF is higher than the degraded broadband electromagnetic distribution image I, further sampling the point spread function to make the sampling rate the same as the degraded broadband electromagnetic distribution image I;

(3) determining an iteration kernel by adopting a maximum likelihood method as follows:

wherein J^t+1For images obtained for the latest iteration, J^tFor the image obtained in the previous iteration, the PSF is a point spread function, and B is a poisson process for image degradation;

(4) iterating the degraded broadband electromagnetic distribution image by using the iteration kernel in the step (3), and carrying out regularity inspection on each result;

(5) and stopping iteration according to the input iteration times, and outputting a recovered image J obtained by the latest iteration.

Images