High-speed target detection method based on planar interferometric optoelectronic imaging system

Abstract

The present invention discloses a high-speed target detection method based on a planar interferometric photoelectric imaging system, the method comprising the following steps: step 1, designing parameters of the planar interferometric photoelectric imaging system; step 2, calculating spectrum acquisition data of the planar interferometric photoelectric imaging system and analyzing its imaging performance; step 3, constructing a detection data model of the sampling spectrum of the planar interferometric photoelectric imaging system; step 4, designing a frequency-space domain target detection algorithm; step 5, comparing and analyzing the target detection speed in the frequency-space domain, and then optimizing the system parameters. The present invention uses the sparse discrete spectrum of the observation scene acquired by the planar interferometric photoelectric imaging system to construct a data model for high-speed target detection, which has the advantages of being able to directly perform target detection in the frequency domain, having a small amount of detection model data, and having a fast target detection speed.

Description

Technical Field

[0001] The present invention belongs to the technical field of interferometric imaging and target detection, and relates to a high-speed target detection method, and in particular to a high-speed frequency domain target detection method based on interferometric imaging and target detection technology. Background Art

[0002] Target detection using high-resolution remote sensing imagery is a research hotspot in remote sensing and plays a crucial role in the military. As the resolution of optoelectronic imaging systems increases, richer and more detailed optical information can be detected. However, this significantly increases the amount of image data, leading to a significant decrease in information processing speed. Therefore, significantly increasing target detection speed while maintaining the resolution of optoelectronic imaging systems has become a pressing issue.

[0003] To significantly reduce the length, volume, and weight of optical imaging systems, the United States proposed a planar interferometric photoelectric imaging system in 2012. This system uses a microlens array to collect light and implements interference pairing, phase adjustment, waveguide coupling, and coherent detection within a photonic integrated circuit. Signal processing yields a spatial image, reducing the length, volume, and weight of the optical system by 10 to 100 times.

[0004] Planar interferometric photoelectric imaging systems sample the observed scene in the frequency domain, and the amount of data obtained is far less than that obtained from spatial imaging at the same resolution. Exploring high-speed frequency-domain target detection methods, leveraging the low frequency-domain data acquisition characteristics of planar interferometric photoelectric imaging systems, has important theoretical and practical value. Summary of the Invention

[0005] The present invention provides a high-speed target detection method based on a planar interferometric photoelectric imaging system. The system uses a sparse discrete frequency spectrum of an observation scene collected by the planar interferometric photoelectric imaging system to construct a data model for high-speed target detection.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A high-speed target detection method based on a planar interferometric photoelectric imaging system comprises the following steps:

[0008] Step 1: Design the parameters of the planar interferometric photoelectric imaging system, including the microlens array arrangement, operating band, object distance, microlens aperture, system aperture, longest baseline size, number of microlenses on each interferometer arm, number of interferometer arms, and number of spectral channels;

[0009] Step 2: Calculate the spectrum acquisition data of the planar interferometric photoelectric imaging system and analyze its imaging performance. The specific steps are as follows:

[0010] Step 21: Calculate the coordinates of each microlens detection unit:

[0011] Using the radial arrangement of microlenses, based on geometric relationships, the radius of the circle enclosed by the innermost layer of the microlens array is calculated as:

[0012]

[0013] Where d is the microlens aperture, α is the angle between two adjacent PICs, and the center coordinates of each microlens are:

[0014]

[0015] Where i represents the number of microlenses, j represents the number of interferometer arms, and x represents the number of lens 、y lens Respectively represent the horizontal and vertical coordinates of the microlens;

[0016] Step 22: Calculate the detection spectrum of each wavelength channel obtained by pairing the microlenses according to the baseline pairing rule:

[0017] According to the baseline pairing rule, the mth microlens and the nth microlens on the jth PIC are paired, and the spatial frequency point (u, v) detected by them is calculated as follows:

[0018]

[0019] Where λ is the operating wavelength and Z is the object distance;

[0020] Step 2: Superimpose the spectrum of each channel;

[0021] Step 24: Perform inverse Fourier transform on the superimposed spectrum to obtain an imaging result;

[0022] Step 3: Construct a detection data model for the sampled spectrum of the planar interferometric photoelectric imaging system. The sampled spectrum modulus matrix contains a large number of zero elements. To eliminate zero elements at non-sampling points, the spectrum modulus of each sampling vector is arranged row by row to form a small spectrum modulus matrix. Using this small spectrum modulus matrix for target detection is faster than using spatial domain images for target detection.

[0023] Step 4: Design a frequency-space domain target detection algorithm based on the combination of directional gradient histogram and support vector machine. The specific steps are as follows:

[0024] Step 41: Given an observation scene dataset, obtain a scene imaging dataset A of a planar interferometric optoelectronic imaging system according to step 2;

[0025] Step 42: Modulo each sampling vector in the spectrum data of data set A and arrange them row by row to form a small matrix data set B;

[0026] Step 43: Perform directional gradient histogram feature extraction on dataset A and dataset B respectively to obtain dataset A1 and dataset B1;

[0027] Step 4: Use support vector machine to perform classification training on dataset A1 and dataset B1 respectively, and calculate the detection time;

[0028] Step 5. Compare and analyze the target detection speed in the frequency and space domain designed in Step 4, and then optimize the system parameters: Modify the design parameters of the planar interferometric optoelectronic imaging system, that is, re-execute Steps 1 to 4 for each given observation scene data set, optimize the system parameters while ensuring the target detection accuracy is higher than 90%, and further improve the detection speed until the speed reaches the highest.

[0029] If other detection algorithms are used, such as the famous YOLO (You Only Look Once) algorithm, step 4 is replaced with:

[0030] Design a frequency-spatial domain target detection algorithm based on YOLO. The specific steps are as follows:

[0031] Step 41: Given an observation scene dataset, obtain a scene imaging dataset A of a planar interferometric optoelectronic imaging system according to step 2;

[0032] Step 42: Modulo each sampling vector in the spectrum data of data set A and arrange them row by row to form a small matrix data set B;

[0033] Step 43: Feature labeling of dataset A and dataset B respectively;

[0034] Step 4. Use the YOLO algorithm to perform classification training and detection on dataset A and dataset B respectively, and calculate the detection time.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] The present invention uses the sparse discrete frequency spectrum of the observation scene collected by the planar interferometric photoelectric imaging system to construct a data model for high-speed target detection, which has the advantages of being able to directly detect targets in the frequency domain, having a small amount of detection model data, and having a fast target detection speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flowchart of the design and imaging simulation program for the planar interferometric optoelectronic imaging system;

[0038] Figure 2 is the light distribution map of the observed scene instance;

[0039] Figure 3This is the sampling spectrum of the planar interferometric photoelectric imaging system;

[0040] Figure 4 is a three-dimensional graph of the sampled spectrum modulus matrix;

[0041] Figure 5 It is an imaging image formed by the planar interferometric photoelectric imaging system on the observed scene;

[0042] Figure 6 Image after constructing the sampled spectrum data model;

[0043] Figure 7 The figure is a flowchart of the specific steps of the high-speed target detection method based on the planar interferometric optoelectronic imaging system. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described below with reference to the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention that does not depart from the spirit and scope of the technical solution of the present invention should be included in the scope of protection of the present invention.

[0045] The present invention provides a high-speed target detection method based on a planar interferometric photoelectric imaging system. Figure 7 As shown, the method includes the following steps:

[0046] Step 1: Design the parameters of the planar interferometric photoelectric imaging system.

[0047] In 2013, the concept of a planar interferometric photoelectric imaging system was proposed in the United States. Recent research has detailed the imaging principles and performance of this system (e.g., Qiuhui Chu et al., Numerical simulation and optimal design of segmented planar imaging detector for electro-optical reconnaissance, Optics Communications, 2017, 405:288–296. http: / / dx.doi.org / 10.1016 / j.optcom.2017.08.021). In a planar interferometric photoelectric imaging system, two microlenses are paired on each interferometer arm to form an interferometric baseline. The light radiation from the observed scene is collected by the microlens array and then passed through the array's band grating to form an ultra-narrowband spectrum. The ultra-narrowband spectrum generated by the microlens pair enters a coupling module, forming interference fringes. The interference fringes are then converted into electrical signals by a photodetector in the coupling module. Finally, the electrical signals are combined into spectral data, which is then subjected to an inverse Fourier transform to obtain an image of the observed scene.

[0048] Here, we present the parameter design results and their significance, along with examples. Table 1 shows the design parameter names, symbols, and values ​​for a planar interferometric optoelectronic imaging system. The physical meanings of these parameters have been extensively discussed in numerous publications (e.g., Qiuhui Chu et al., Numerical simulation and optimal design of segmented planar imaging detector for electro-optical reconnaissance, Optics Communications, 2017, 405:288–296, http: / / dx.doi.org / 10.1016 / j.optcom.2017.08.021). The parameter settings and symbols used here follow those from previous literature. This design example is for a remote sensing imaging system, so the object distance Z is large, and l and p also increase to a certain extent.

[0049] Table 1

[0050]

[0051]

[0052] Step 2: Calculate the spectrum acquisition data of the planar interferometric optoelectronic imaging system and analyze its imaging performance.

[0053] The flow chart of the imaging design simulation program of the planar interferometric optoelectronic imaging system is as follows: Figure 1 As shown, the simulation steps include inputting system parameters (microlens array arrangement, operating wavelength, object distance, microlens aperture, system aperture, longest baseline size, number of microlenses on each interferometer arm, number of interferometer arms, and number of spectral channels), calculating the coordinates of each microlens detection unit, calculating the detection spectrum of each wavelength channel obtained by pairing microlenses according to the baseline pairing rule, then superimposing the spectra of each channel, and finally performing an inverse Fourier transform on the superimposed spectrum to obtain the imaging result. Using the radial arrangement of microlenses, based on the geometric relationship, the radius of the circle enclosed by the innermost layer of the microlens array can be calculated as:

[0054]

[0055] Where d is the microlens aperture, and α is the angle between two adjacent PICs. The center coordinates of each microlens are:

[0056]

[0057] Where i represents the number of microlenses, j represents the number of interferometer arms, and x represents the number of lens 、y lensRepresent the horizontal and vertical coordinates of the microlens respectively. According to the baseline pairing rule, the m-th microlens and the n-th microlens on the j-th PIC are paired, and the calculation method of the spatial frequency point (u, v) detected by them is:

[0058]

[0059] Where λ is the operating wavelength and Z is the object distance.

[0060] An example of light distribution in the observed scene is Figure 2 As shown, the microlens array adopts the uniform pairing rule. According to formula (3), the value of the spatial frequency point (u, v) can be calculated, which can be obtained as follows: Figure 3 The sample spectrum diagram of the planar interferometric optoelectronic imaging system is shown, where different colors correspond to different spectral channels. Figure 3 The corresponding matrix dimension is 300×300. Figure 3 The corresponding matrix modulo can be obtained Figure 4 The dimension shown is 300×300 of the sample spectrum matrix. Figure 3 The inverse Fourier transform of the spectrum shown can be obtained Figure 5 The reconstructed image shown has a pixel data of a 300×300 matrix. Figure 5 That is Figure 2 The image of the observed scene formed by the plane interferometric photoelectric imaging system can be seen from the sparse sampling of the plane interferometric photoelectric imaging system. Figure 5 Blurring and degradation of the image shown.

[0061] Step 3: Construct a detection data model of the sampling spectrum of the planar interferometric photoelectric imaging system.

[0062] Figure 4 There are a lot of zero elements in the sampled spectrum modulus matrix shown. To eliminate the zero elements of non-sampling points, the spectrum modulus of each sampling vector is arranged row by row to form a small matrix of 129×38. Figure 6 shown. Figure 6 The amount of data only accounts for Figure 5 5.45% of the data volume, so Figure 6 The speed of target detection by the spectrum mode data model shown is faster than that of Figure 5 The speed of object detection in the spatial domain image shown is high.

[0063] Step 4: Design a frequency-spatial domain target detection algorithm.

[0064] There are many types of target detection algorithms. This paper only uses the detection algorithm based on the combination of directional gradient histogram and support vector machine as an example to illustrate. Figure 2Similar satellite images are used as observation scene datasets, and the number of datasets is no less than 300. Figure 1 The process shown obtains the scene imaging data set A of the planar interferometric optoelectronic imaging system. Then, each sampling vector in the spectrum data of data set A is modulo and arranged row by row to form a small matrix data set B of 129×38. Directed gradient histogram feature extraction is performed on data set A and data set B respectively, and data set A1 and data set B1 can be obtained respectively. Then, support vector machine is used to perform classification training on data set A1 and data set B1 respectively, and the detection time is counted. The present invention only describes the target detection algorithm based on the combination of directional gradient histogram and support vector machine as an example. Other detection algorithms can also be used, but the target detection algorithm used must be able to simultaneously perform target detection on the spatial-frequency domain model.

[0065] Step 5: Compare and analyze the detection speed in the frequency and space domains, and then optimize the system parameters.

[0066] Repeating step 4, constructing datasets of different scene samples, and conducting extensive simulation analysis and algorithm research, the team concluded that, while maintaining a target detection accuracy greater than 92%, frequency-domain target detection speed is 5 to 8 times faster than spatial-domain target detection speed. However, this result is not optimal, as the improvement in frequency-domain detection speed depends on the characteristics of the observation scene. To further improve the frequency-domain detection speed and achieve the highest possible frequency-domain target detection speed, the design parameters of the planar interferometric optoelectronic imaging system need to be modified. Specifically, steps 1 to 4 are repeated for each given observation scene dataset, optimizing the system parameters while maintaining a target detection accuracy greater than 90% to further increase the detection speed until the maximum speed is reached.

[0067] A desktop computer with a CPU main frequency of 3.6GHz (4 cores, 8 threads) and 8G memory is used for spatial and frequency domain target detection. When the detection accuracy is higher than 96%, the time consumption results are: spatial domain detection takes 87.1271 milliseconds and frequency domain detection takes 12.5463 milliseconds. The frequency domain detection speed is increased to 6.94 times the speed of spatial domain detection.

Claims

1. A high-speed target detection method based on a planar interferometric photoelectric imaging system, characterized in that The method comprises the following steps: Step 1: Design the parameters of the planar interferometric photoelectric imaging system, including the microlens array arrangement, operating band, object distance, microlens aperture, system aperture, longest baseline size, number of microlenses on each interferometer arm, number of interferometer arms, and number of spectral channels; Step 2: Calculate the spectrum acquisition data of the planar interferometric photoelectric imaging system and analyze its imaging performance; Step 3: Construct a detection data model for the sampled spectrum of the planar interferometric photoelectric imaging system. The specific steps are as follows: There are a large number of zero elements in the sampled spectrum modulus matrix. To eliminate the zero elements at non-sampling points, the spectrum modulus of each sampling vector is arranged row by row to form a small spectrum modulus matrix. The speed of target detection using the small spectrum modulus matrix is ​​higher than that of target detection using spatial domain images. Step 4: Design a frequency-space domain target detection algorithm based on the combination of directional gradient histogram and support vector machine. The specific steps are as follows: Step 41: Given an observation scene dataset, obtain a scene imaging dataset A of a planar interferometric optoelectronic imaging system according to step 2; Step 42: Modulo each sampling vector in the spectrum data of data set A and arrange them row by row to form a small matrix data set B; Step 43: Perform directional gradient histogram feature extraction on dataset A and dataset B respectively to obtain dataset A1 and dataset B1; Step 4: Use support vector machine to perform classification training on dataset A1 and dataset B1 respectively, and calculate the detection time; Step 5: Compare and analyze the target detection speed in the frequency and space domain designed in step 4, and then optimize the system parameters.

2. The high-speed target detection method based on a planar interferometric photoelectric imaging system according to claim 1, characterized in that The specific steps of step 2 are as follows: Step 21: Calculate the coordinates of each microlens detection unit: Using the radial arrangement of microlenses, based on geometric relationships, the radius of the circle enclosed by the innermost layer of the microlens array is calculated as: Where d is the microlens aperture, α is the angle between two adjacent PICs, and the center coordinates of each microlens are: Where i represents the number of microlenses, j represents the number of interferometer arms, and x represents the number of lens 、y lens Respectively represent the horizontal and vertical coordinates of the microlens; Step 22: Calculate the detection spectrum of each wavelength channel obtained by pairing the microlenses according to the baseline pairing rule: According to the baseline pairing rule, the mth microlens and the nth microlens on the jth PIC are paired, and the spatial frequency point (u, v) detected by them is calculated as follows: Where λ is the operating wavelength and Z is the object distance; Step 2: Superimpose the spectrum of each channel; Step 24: Perform inverse Fourier transform on the superimposed spectrum to obtain the imaging result.

3. The high-speed target detection method based on a planar interferometric photoelectric imaging system according to claim 1, characterized in that The above step 4 is replaced by: designing a frequency-spatial domain target detection algorithm based on YOLO. The specific steps are as follows: Step 41: Given an observation scene dataset, obtain a scene imaging dataset A of a planar interferometric optoelectronic imaging system according to step 2; Step 42: Modulo each sampling vector in the spectrum data of data set A and arrange them row by row to form a small matrix data set B; Step 43: Feature labeling of dataset A and dataset B respectively; Step 4. Use the YOLO algorithm to perform classification training and detection on dataset A and dataset B respectively, and calculate the detection time.

4. The high-speed target detection method based on a planar interferometric photoelectric imaging system according to claim 1, characterized in that The specific steps of step five are as follows: modify the design parameters of the planar interferometric optoelectronic imaging system, that is, re-execute steps one to four for each given observation scene data set, optimize the system parameters while ensuring the target detection accuracy rate is higher than 90%, and further improve the detection speed until the maximum speed is reached.

Images