Multi-frame association long-distance dynamic small target identification method based on frequency domain enhancement

Abstract

The invention discloses a multi-frame association long-distance dynamic small target identification method based on frequency domain enhancement, which relates to the field of image processing, and is characterized in that a frequency domain imaging principle and a multi-frame association method are combined, and a time domain image is converted to a frequency domain by using an image sequence acquired by a single photon counting camera Photon Force PF32, so that a target identification result is obtained; the method comprises the following steps: effectively distinguishing the spectrum characteristics of a small target from the spectrum characteristics of noise and background interference, suppressing the noise and background interference through Fourier transform, introducing a multi-frame association method, and analyzing a multi-frame image after frequency domain processing to obtain a motion track and a characteristic change rule of the small target between continuous frames; therefore, a clearer target is obtained through shift accumulation. Through the method, the advantage of high sensitivity of the single-photon counting camera is brought into full play, the problem that the false detection probability of the long-distance dynamic weak and small target is large is effectively solved, and the accuracy and reliability of target detection are improved.

Description

Technical Field

[0001] The invention relates to the field of image processing, and in particular to a multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement. Background Art

[0002] With the evolution of computer vision technology, target detection has gradually expanded from basic research to multimodal application scenarios. Different from conventional target detection tasks, small target detection has become a key research topic in the field of computer vision due to its important application value in military reconnaissance, remote sensing monitoring, smart transportation, medical image analysis and other fields. Especially in battlefield perception systems, subject to the development of stealth technology and the physical limitations of sensors, long-distance dynamic small target detection has become a core challenge for modern defense systems.

[0003] Current detection technology faces three core challenges:

[0004] First, the similarity of the spectral characteristics of small targets and noise in low signal-to-noise ratio environments causes the failure of traditional single-frame detection methods;

[0005] Second, complex background interference causes a significant increase in false alarm rate;

[0006] Third, the real-time requirements of mobile platforms conflict with the limitations of computing resources. Experimental data show that when the target pixel ratio is less than 0.12%, the recall rate of mainstream detection models drops by more than 60%, which seriously restricts the practical application in complex environments with small targets. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement, which solves the problem that the probability of false detection of long-distance dynamic weak small targets is high under traditional methods.

[0008] To achieve the above object, the present invention provides a multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement, comprising the following steps:

[0009] Step S1: Use a single-photon counting camera, Photon Force PF32, to collect data of small targets in a low signal-to-noise ratio environment, effectively capture weak light signals, and provide a basis for raw data;

[0010] Step S2: In different scenes, a continuous image is randomly extracted, and a spectrum analysis is performed on a single pixel point in the small target for the extracted continuous image to generate corresponding spectrum data, realize the conversion from time domain to frequency domain, and mine the feature information of the image in the frequency dimension;

[0011] Step S3: Perform Fourier transform on each pixel in the image, use the advantages of frequency domain processing to improve the signal-to-noise ratio of the image, and extract the target with enhanced energy from the original low signal-to-noise ratio image by analyzing and processing the spectrum data, highlighting the characteristic performance of the target in the frequency domain;

[0012] Step S4: using a specific program to estimate the approximate speed of the target movement, the estimation process is based on the acquired target data and related motion characteristics, providing important motion parameter references for subsequent multi-frame association processing;

[0013] Step S5: Implement fixed frame interval grouping operation and perform related imaging, perform related imaging on the small target data collected in step S1, and then divide it into several groups according to fixed frame intervals to achieve preliminary sorting and organization of the data;

[0014] Step S6: carrying out multi-hypothesized speed multi-frame association traversal, for the input multi-frame image sequence, performing shift accumulation operations on the images according to different hypothesized speeds to generate multiple trailing target images;

[0015] Step S7: performing a multi-frame associated shift accumulation operation on the target, and performing shift and accumulation processing on the target based on the multi-frame images to enhance the target signal and suppress noise and background interference;

[0016] Step S8: Obtain a small target image sequence with a high signal-to-noise ratio.

[0017] Preferably, in step S3, the specific steps of performing Fourier transform on all pixel points are:

[0018] Step 1: Traverse each row of the image, obtain the time series of all pixels in each row, and reshape it into a matrix with b rows and a columns, where each column represents the time series of a pixel;

[0019] Step 2: Traverse each column of the current row, obtain the time series for each pixel, remove the zero values, and find the index of non-zero data;

[0020] Step 3: Convert the index of non-zero data into time units.

[0021] Step 4: Use the formula X=abs(sum(exp(-1j*t*2*pi*fm))) to calculate the Fourier transform amplitude of the pixel at the target frequency fm analyzed in step S2, and store the result in the matrix Y;

[0022] Step 5: Display the normalized Fourier transform result image.

[0023] Preferably, in step S4, the specific steps for estimating the approximate speed of the target movement are: after completing the Fourier transform, if the target appears to be roughly resolved in the transformed image, select the central pixel point of the target on the first and last images in the sequence, and make a preliminary estimate of the direction of the target movement speed, Vx = the number of pixels / frames in the x direction, Vy = the number of pixels / frames in the y direction.

[0024] Preferably, in step S5, the specific steps of the fixed frame interval grouping are as follows:

[0025] Step S51: define a fixed frame interval, where the interval represents the number of frames contained in each set of data;

[0026] Step S52: Calculate the number of groups according to the total number of frames c and the fixed frame interval, and create a cell array for storing the grouped data;

[0027] Step S53: by looping through each group, determining the start frame index and the end frame index of the group, and storing the corresponding frame data in a cell array, completing the fixed frame interval grouping operation;

[0028] Step S54: Divide the original small target data into several groups according to fixed frame intervals.

[0029] Preferably, in step S5, the specific steps of the relevant imaging operation are as follows:

[0030] S5-1: Create an all-zero matrix Y for storing correlation results between adjacent frames;

[0031] S5-2: By looping through every two adjacent frames, subtracting the average background from each frame data and performing element-by-element multiplication, the correlation result between adjacent frames is obtained and stored in the Y matrix;

[0032] S5-3: sum the Y matrix in the frame dimension and divide it by the total number of frames c to obtain an average correlation result;

[0033] S5-4: Normalize the average correlation result so that its value range is between [0,1].

[0034] Preferably, in step S6, the specific steps of carrying out multi-hypothesis speed multi-frame association traversal are:

[0035] Step S61: when the real moving speed of the target is in an unknown state, n assumed speeds of the target are manually set;

[0036] Step S62: The n assumed velocities in step 61 are expressed in the form of a two-dimensional vector, (V x1 ,V y1 )、(V x2 ,V y2)……(V xn ,V yn ), where V x represents the target's assumed velocity component in the horizontal direction, V y Represents the hypothetical velocity component of the target in the vertical direction.

[0037] Preferably, in step S7, the specific steps of performing multi-frame associated shift accumulation on the target are:

[0038] Step S71: performing multi-frame shift accumulation processing on the input multi-frame image sequence according to the assumed speed set in step S62;

[0039] S72: For the i-th assumed speed (V xi ,V yi ), performing a corresponding shift operation on each frame of the multi-frame image sequence according to the speed, and during the shifting process, the image frame moves in the direction and magnitude indicated by the assumed speed;

[0040] S73: After completing the shift of each frame of the image, the target's real speed and moving direction are determined based on the length of the tail, and then these shifted images are accumulated to integrate the information in multiple frames of images, enhance the target feature expression, and mine the target's movement traces from the comprehensive information of multiple frames of images.

[0041] Therefore, the present invention adopts the above-mentioned multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement, which has the following beneficial effects:

[0042] (1) The present invention organically combines the frequency domain imaging principle with the multi-frame association method to effectively distinguish the spectral characteristics of small targets from noise and background interference, greatly reducing the probability of false detection of long-distance dynamic weak targets, and providing more reliable target detection results for military, remote sensing and other fields.

[0043] (2) The present invention performs Fourier transform on the image pixel by pixel and performs multi-frame associated shift accumulation operations, which can fully explore the characteristic information of the image in the frequency dimension, enhance the target signal, suppress noise and background interference, and stably detect small targets even in a low signal-to-noise ratio environment, thus solving the problem that the traditional single-frame detection method fails in this environment.

[0044] (3) The present invention adopts a unique target speed estimation method. By selecting the central pixel points of the target on the first and last two images in the sequence, it can make a preliminary estimate of the target movement speed direction after Fourier transform, providing key motion parameter references for subsequent multi-frame association processing, which helps to track the target movement trajectory more accurately.

[0045] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a method for recognizing small dynamic targets at a long distance based on multi-frame association and frequency domain enhancement according to the present invention;

[0047] Figure 2 A small target image with low signal-to-noise ratio in an embodiment of the present invention;

[0048] Figure 3 The image after Fourier transformation pixel by pixel in the embodiment of the present invention;

[0049] Figure 4 is a schematic diagram of estimating the true speed in an embodiment of the present invention;

[0050] Figure 5 is the correlation imaging result in the embodiment of the present invention;

[0051] Figure 6 It is a fixed frame interval grouping operation in an embodiment of the present invention;

[0052] Figure 7 It is a result image of multi-hypothesis speed multi-frame association traversal and shift alignment accumulation in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0054] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0055] The words "include" or "comprises" and the like used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms "inside", "outside", "upper", "lower", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly specified and limited, the terms "attachment" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] Example

[0057] like Figure 1 As shown, a multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement includes the following steps:

[0058] Step S1: Use a single-photon counting camera, Photon Force PF32, to collect data of small targets in low signal-to-noise ratio environments, such as at night or in foggy weather. Figure 2 As shown, weak light signals are effectively captured to provide a basis for raw data; a small UAV is prepared for target simulation, and the UAV is kept at a relatively far distance so that it appears as a small target in the image.

[0059] Step S2: In different scenarios, a continuous image is randomly extracted. For the extracted continuous image, a spectrum analysis is performed on a single pixel point that may exist in the small target to generate corresponding spectrum data, realize the conversion from time domain to frequency domain, and mine the feature information of the image in the frequency dimension.

[0060] Step S3: Perform Fourier transform on each pixel in the image, use the advantages of frequency domain processing to improve the signal-to-noise ratio of the image, and extract the target with enhanced energy from the original low signal-to-noise ratio image by analyzing and processing the spectrum data, highlighting the characteristic performance of the target in the frequency domain;

[0061] For Fourier transform, the target frequency fm is set to 533 Hz. This frequency is calculated based on the previous analysis of the spectrum characteristics of similar small targets, which helps to highlight the characteristics of the target in the frequency domain.

[0062] fm = 8000 / 60*4 = 533.3 Hz rotation per minute / 60s*number of rotors;

[0063] In step S2 and step S3, the specific steps of performing Fourier transform on each pixel are as follows:

[0064] Step 1: Traverse each row of the image, obtain the time series of all pixels in each row, and reshape it into a matrix with b rows and a columns, where each column represents the time series of a pixel;

[0065] Step 2: Traverse each column of the current row, obtain the time series for each pixel, remove the zero values, and find the index of non-zero data;

[0066] Step 3: Convert the index of non-zero data into time units.

[0067] Step 4: Use the formula X=abs(sum(exp(-1j*t*2*pi*fm))) to calculate the Fourier transform amplitude of the pixel at the target frequency fm analyzed in step S2, and store the result in the matrix Y;

[0068] Step 5: Display the normalized Fourier transform result image

[0069] Step S4: using a specific program to estimate the approximate speed of the target movement, the estimation process is based on the acquired target data and related motion characteristics, providing important motion parameter references for subsequent multi-frame association processing;

[0070] In step S4, the specific steps for estimating the approximate speed of the target movement are as follows: after completing the Fourier transform, if the target appears to be roughly resolved in the transformed image, select the central pixel point of the target on the first and last images in the sequence, and make a preliminary estimate of the direction of the target movement speed, Vx = the number of pixels / frames in the x direction, Vy = the number of pixels / frames in the y direction.

[0071] Step S5: Implement the fixed frame interval grouping operation and perform related imaging, and set the fixed frame interval to 10,000 frames, that is, each group of data contains 10,000 frames of images. This interval is determined through multiple experiments, taking into account the target movement speed and image information integrity. The small target data collected in step S1 is divided into several groups according to the fixed frame number interval to achieve preliminary sorting and organization of the data;

[0072] In step S5, the specific steps of framing interval grouping are as follows:

[0073] Step S51: define a fixed frame interval, where the interval represents the number of frames contained in each set of data;

[0074] Step S52: Calculate the number of groups according to the total number of frames c and the fixed frame interval, and create a cell array for storing the grouped data;

[0075] Step S53: by looping through each group, determining the start frame index and the end frame index of the group, and storing the corresponding frame data in a cell array, completing the fixed frame interval grouping operation;

[0076] Step S54: Divide the original small target data into several groups according to fixed frame intervals.

[0077] In step S5, the specific steps of the relevant imaging operation are as follows:

[0078] S5-1: Create an all-zero matrix Y for storing correlation results between adjacent frames;

[0079] S5-2: By looping through every two adjacent frames, subtracting the average background from each frame data and performing element-by-element multiplication, the correlation result between adjacent frames is obtained and stored in the Y matrix;

[0080] S5-3: sum the Y matrix in the frame dimension and divide it by the total number of frames c to obtain an average correlation result;

[0081] S5-4: Normalize the average correlation result so that its value range is between [0,1].

[0082] Step S6: carrying out multi-hypothesized speed multi-frame association traversal, for the input multi-frame image sequence, performing shift accumulation operations on the images according to different hypothesized speeds to generate multiple trailing target images;

[0083] In step S6, in the multi-hypothesized speed multi-frame association traversal, five hypothesized speeds are set, namely (V x1 ,V y1 )、(V x2 ,V y2 )……(V xn ,V yn ) in pixels / frame. These assumed speeds cover a variety of possible motion directions and speed ranges of the target. The specific steps for carrying out multi-hypothesized speed multi-frame association traversal are:

[0084] Step S61: when the real moving speed of the target is in an unknown state, n assumed speeds of the target are manually set;

[0085] Step S62: The n assumed velocities in step 61 are expressed in the form of a two-dimensional vector, (V x1 ,V y1 )、(V x2 ,V y2 )……(V xn ,V yn ), where V x represents the target's assumed velocity component in the horizontal direction, V y Represents the hypothetical velocity component of the target in the vertical direction.

[0086] Step S7: performing a multi-frame associated shift accumulation operation on the target, and performing shift and accumulation processing on the target based on the multi-frame images to enhance the target signal and suppress noise and background interference;

[0087] In step S7, the specific steps of performing multi-frame associated shift accumulation on the target are:

[0088] Step S71: performing multi-frame shift accumulation processing on the input multi-frame image sequence according to the assumed speed set in step S62;

[0089] S72: For the i-th assumed speed (V xi ,V yi), performing a corresponding shift operation on each frame of the multi-frame image sequence according to the speed, and during the shifting process, the image frame moves in the direction and magnitude indicated by the assumed speed;

[0090] S73: After completing the shift of each frame of the image, the target's real speed and moving direction are determined based on the length of the tail, and then these shifted images are accumulated to integrate the information in multiple frames of images, enhance the target feature expression, and mine the target's movement traces from the comprehensive information of multiple frames of images.

[0091] Step S8: Obtain a small target image sequence with a high signal-to-noise ratio.

[0092] Experimental procedures

[0093] Data collection

[0094] In the simulation environment, the small UAV was controlled to fly at different speeds and trajectories. The single-photon counting camera Photon Force PF32 was used to collect image sequences of the UAV flying in a low signal-to-noise ratio environment to obtain raw image data.

[0095] The pixel-by-pixel Fourier transform is Figure 3 As shown;

[0096] 1. Follow the detailed steps of S3 in the technical solution to traverse each row of the image. For each row, obtain the time series of all pixels in the row and reshape it into a matrix with b rows and a columns, where each column represents the time series of a pixel.

[0097] 2. In the inner loop, traverse each column of the current row, obtain the time series of each pixel, remove the zero values, find the index of the non-zero data, and convert it to time units (first to picoseconds, then to seconds).

[0098] 3. Use the formula to calculate the Fourier transform amplitude of the pixel at the target frequency fm, and store the result in the matrix Y. Display the normalized Fourier transform result image, and observe the amplitude information of each pixel at the target frequency. At this time, it can be seen that the characteristics of the target in the frequency domain are enhanced and distinguished from noise and background interference.

[0099] The target speed estimation result is as follows: Figure 4 As shown;

[0100] After the Fourier transform is completed, if the target is roughly distinguishable in the transformed image, select the central pixel point of the target in the first and last images in the sequence. According to the formula Vx = number of pixels in the x direction / number of frames, Vy = number of pixels in the y direction / number of frames, preliminarily estimate the direction of the target's motion speed. For example, in a certain experiment, the number of pixels in the x direction and the number of pixels in the y direction of the target center pixel point in the first and last images is 20, and the number of frames is 100, so it is estimated that Vx = 0.2 pixels / frame, Vy = 0.1 pixels / frame.

[0101] The fixed frame interval grouping operation and related imaging results are as follows Figure 5 and Figure 6 As shown;

[0102] 1. Calculate the number of groups according to the set fixed frame interval of 10,000 frames. Create a cell array to store the grouped data. Loop through each group, determine the start frame index and end frame index of each group, and store the corresponding frame data in the cell array to complete the fixed frame interval grouping operation.

[0103] 2. Perform correlation imaging processing, taking the correlation of adjacent frames as an example. Create an all-zero matrix Y to store the correlation results between adjacent frames. Loop through each two adjacent frames, subtract the average background from each frame data, and multiply element by element to obtain the correlation results between adjacent frames and store them in the Y matrix. Then, sum the Y matrix in the third dimension (frame dimension) and divide it by the total number of frames c to obtain the average correlation result. Finally, normalize the average correlation result so that its value range is between [0,1].

[0104] The results of multi-hypothesis speed multi-frame association traversal and shift accumulation are as follows Figure 7 As shown;

[0105] 1. For the five assumed speeds, perform multi-frame shift accumulation processing on the input multi-frame image sequence. For the i-th assumed speed (Vxi, Vyi), perform corresponding shift operations on each frame image in the multi-frame image sequence according to this speed. For example, for the assumed speed (Vxi, Vyi) = (2, 1), a frame image moves 2 pixel units in the horizontal direction and 1 pixel unit in the vertical direction.

[0106] 2. After completing the shift of each frame of the image, the target's true speed and direction of movement are determined based on the length of the tail. The true speed diagram is as follows: Figure 4 As shown; these shifted images are accumulated to integrate the information in multiple frames of images and enhance the feature expression of the target.

[0107] Experimental Results

[0108] After the above series of processing steps, a small target image sequence with a high signal-to-noise ratio is obtained. Compared with the low signal-to-noise ratio image originally collected, the target is clearer in the processed image and the features are more obvious. Through comparative experiments, the detection recall rate of drone targets under the same simulation environment using the traditional single-frame detection method is only 30%, while the detection recall rate of the method of the present invention reaches 85%, which effectively reduces the false detection probability of long-distance dynamic weak targets, improves the accuracy and reliability of target detection, and verifies the effectiveness of the technical solution of the present invention.

[0109] Therefore, the present invention adopts the above-mentioned multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement, and uses Fourier transform and multi-frame association methods to analyze the multi-frame images after frequency domain processing, and obtains the motion trajectory and feature change law of the small target between consecutive frames, thereby shifting and accumulating to obtain a clearer target.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-frame association long-distance dynamic small target recognition method based on frequency domain enhancement, characterized by: The following steps are involved: Step S1: Use a single-photon counting camera, Photon Force PF32, to collect data of small targets in a low signal-to-noise ratio environment, effectively capture weak light signals, and provide a basis for raw data; Step S2: In different scenes, a continuous image is randomly extracted, and a spectrum analysis is performed on a single pixel point in the small target for the extracted continuous image to generate corresponding spectrum data, realize the conversion from time domain to frequency domain, and mine the feature information of the image in the frequency dimension; Step S3: Perform Fourier transform on all pixels in the image, use the advantages of frequency domain processing to improve the signal-to-noise ratio of the image, and extract the energy-enhanced target from the original low-signal-to-noise ratio image by analyzing and processing the spectrum data, highlighting the characteristic performance of the target in the frequency domain; Step S4: using a specific program to estimate the approximate speed of the target movement, the estimation process is based on the acquired target data and related motion characteristics, providing important motion parameter references for subsequent multi-frame association processing; Step S5: Implement fixed frame interval grouping operation and perform related imaging, perform related imaging on the small target data collected in step S1, and then divide it into several groups according to fixed frame intervals to achieve preliminary sorting and organization of the data; Step S6: carrying out multi-hypothesized speed multi-frame association traversal, for the input multi-frame image sequence, performing shift accumulation operations on the images according to different hypothesized speeds to generate multiple trailing target images; Step S7: performing a multi-frame associated shift accumulation operation on the target, and performing shift and accumulation processing on the target based on the multi-frame images to enhance the target signal and suppress noise and background interference; Step S8: Obtain a small target image sequence with a high signal-to-noise ratio.

2. According to the method for recognizing small dynamic targets at long distances based on multi-frame association and frequency domain enhancement described in claim 1, it is characterized by: In step S3, the specific steps of performing Fourier transform on each pixel are as follows: Step 1: Traverse each row of the image, obtain the time series of all pixels in each row, and reshape it into a matrix with b rows and a columns, where each column represents the time series of a pixel; Step 2: Traverse each column of the current row, obtain the time series for each pixel, remove the zero values, and find the index of non-zero data; Step 3: Convert the index of non-zero data into time units; Step 4: Use the formula X = abs( sum ( exp (-1j*t*2*pi*fm))) calculates the Fourier transform amplitude of the pixel at the target frequency fm analyzed in step S2, and stores the result in matrix Y; Step 5: Display the normalized Fourier transform result image.

3. The method for recognizing small dynamic targets at long distances based on multi-frame association and frequency domain enhancement according to claim 1, characterized in that: In step S4, the specific steps for estimating the approximate speed of the target movement are as follows: after completing the Fourier transform, if the target appears to be roughly resolved in the transformed image, select the central pixel point of the target on the first and last images in the sequence, and make a preliminary estimate of the direction of the target movement speed, Vx = the number of pixels / frames in the x direction, Vy = the number of pixels / frames in the y direction.

4. The method for recognizing small dynamic targets at long distances based on multi-frame association and frequency domain enhancement according to claim 1, characterized in that: In step S5, the specific steps of framing interval grouping are as follows: Step S51: define a fixed frame interval, where the interval represents the number of frames contained in each set of data; Step S52: Calculate the number of groups according to the total number of frames c and the fixed frame interval, and create a cell array for storing the grouped data; Step S53: by looping through each group, determining the start frame index and the end frame index of the group, and storing the corresponding frame data in a cell array, completing the fixed frame interval grouping operation; Step S54: Divide the original small target data into several groups according to fixed frame intervals.

5. A method for recognizing small dynamic targets at long distances based on multi-frame association and frequency domain enhancement according to claim 4, characterized in that: In step S5, the specific steps of the relevant imaging operation are as follows: S5-1: Create an all-zero matrix Y for storing correlation results between adjacent frames; S5-2: By looping through every two adjacent frames, subtracting the average background from each frame data and performing element-by-element multiplication, the correlation result between adjacent frames is obtained and stored in the Y matrix; S5-3: sum the Y matrix in the frame dimension and divide it by the total number of frames c to obtain an average correlation result; S5-4: Normalize the average correlation result so that its value range is between [0,1].

6. The method for recognizing small dynamic targets at long distances based on multi-frame association and frequency domain enhancement according to claim 1, characterized in that: In step S6, the specific steps of carrying out multi-hypothesis speed multi-frame association traversal are: Step S61: when the real moving speed of the target is in an unknown state, n assumed speeds of the target are manually set; Step S62: The n assumed velocities in step 61 are expressed in the form of a two-dimensional vector, (V x1 ,V y1 )、(V x2 ,V y2 )……(V xn ,V yn ), where V x represents the target's assumed velocity component in the horizontal direction, V y Represents the hypothetical velocity component of the target in the vertical direction.

7. The method for recognizing small dynamic targets at long distances based on multi-frame association and frequency domain enhancement according to claim 6, characterized in that: In step S7, the specific steps of performing multi-frame associated shift accumulation on the target are: Step S71: performing multi-frame shift accumulation processing on the input multi-frame image sequence according to the assumed speed set in step S62; S72: For the i-th assumed speed (V xi ,V yi ), performing a corresponding shift operation on each frame of the multi-frame image sequence according to the speed, and during the shifting process, the image frame moves in the direction and magnitude indicated by the assumed speed; S73: After completing the shift of each frame of the image, the target's real speed and direction of movement are determined based on the length of the tail, and then these shifted images are accumulated to integrate the information in multiple frames of images, enhance the target feature expression, and mine the target's movement traces from the comprehensive information of multiple frames of images.

Images