Multi-hypothesis speed weak and small target detection method based on gyroscope

Abstract

The invention discloses a gyroscope-based multi-hypothesis speed weak and small target detection method, which comprises the following steps of: firstly, performing multi-speed hypothesis accumulation and differential processing on an image sequence under the same coordinate axis based on three-axis information of a gyroscope, eliminating background information on a differential image, detecting a line target, and obtaining target trailing line information; the real speed of the target is determined in combination with line information, the position of the target is reversely deduced, weak and small target detection is performed by using a line correlation contrast algorithm based on centroid extraction, and the background is suppressed in numerous backgrounds and clutters to improve the target contrast. According to the scheme, multi-hypothesis speed accumulation is carried out by means of three-axis information of the gyroscope in the weak and small target detection process, false alarms caused by movement of the small target and background points in the image due to movement of the camera can be avoided, the method can be used in the scene that the camera shakes, and meanwhile under the condition of the low signal-to-noise ratio, the detection accuracy is improved. The method effectively improves the signal-to-noise ratio of the moving small target, inhibits background clutter and strong clutter, improves the detection accuracy, and is especially suitable for the detection of the moving small target under the condition of low signal-to-noise ratio.

Description

Technical Field

[0001] The invention belongs to the field of small target detection, and in particular relates to a small target detection method based on multiple hypothetical velocities of a gyroscope. Background Art

[0002] Infrared small target detection plays an important role in infrared imaging systems and is widely used in early warning, precision guidance, remote sensing and other fields. Infrared search has been a hot research topic in the past few decades. Sometimes due to the application environment, our imaging system is far away from the target. When imaging at a long distance, the target occupies very little space and the signal-to-noise ratio is extremely low, which brings many problems to the detection of weak targets. In addition, due to the movement of the target, the small target is always immersed in a complex background, and the background is disturbed by high-intensity clouds, sharp edges, etc. Therefore, it is a difficult task to accurately detect the position of weak targets.

[0003] At present, the main solutions to this problem are divided into two categories: single-frame image detection and multi-frame image detection. Typical representatives of single-frame image detection include background suppression through spatial filtering, human visual system, HVS method, deep learning-based method and sparsity and low-quality reconstruction-based method.

[0004] Single-frame detection will generate a large number of false alarm points when the signal-to-noise ratio is low. Currently, most multi-frame detection methods, such as multi-level hypothesis detection method, dynamic programming method, three-dimensional filtering detection algorithm, pipeline filtering method and energy accumulation method, process continuous multi-frame images according to the continuity and regularity of the target's movement in the airspace. Dynamic programming and energy accumulation will directly cause energy diffusion to energy accumulation, and it will be difficult to detect targets with large movements, resulting in more false alarm points. The multi-level hypothesis method makes energy accumulation as effective as possible through assumptions and decisions, but it is often applied to simple backgrounds and is difficult to effectively detect when the target is immersed in a complex background.

[0005] In summary, when searching for small and weak targets at long distances, the existing technology will miss the target when the target signal-to-noise ratio is lower than the algorithm detectable signal-to-noise ratio due to the small target size, low radiation intensity and long detection distance. Traditional noise reduction or enhancement algorithms are difficult to overcome the energy dispersion problem caused by target motion. Summary of the invention

[0006] In view of the above problems, the object of the present invention is to provide a small target detection method based on multiple hypothesized velocities of a gyroscope.

[0007] The specific technical solution for achieving the purpose of the present invention is:

[0008] A small target detection method based on multi-hypothesis velocity of gyroscope includes the following steps:

[0009] Step 1: Based on the gyroscope three-axis information, the optical axis model is calculated by multi-hypothesis speed, the continuous images on the time axis correspond to a coordinate system, the image sequence under the same coordinate axis is accumulated by multi-speed hypothesis, and the target is decelerated to improve the target signal-to-noise ratio;

[0010] Step 2: Perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target trailing line information;

[0011] Step 3: extract the line target, calculate the length and slope of the trailing line, and use the line information to calculate the target's true speed;

[0012] Step 4: Use the real motion information of the target to perform reverse superposition, infer the target position, and suppress strong clutter;

[0013] Step 5: Use the line correlation contrast algorithm based on centroid extraction to detect small targets, suppress the background and improve the target contrast among many backgrounds and clutter.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] (1) The solution of the present invention uses the three-axis information of the gyroscope to perform multi-hypothesis velocity accumulation during the small target detection process, which can avoid false alarms caused by the movement of small targets and background points in the image caused by camera motion. It can be used in scenes where the camera is shaking, thereby improving the detection accuracy.

[0016] (2) The solution of the present invention can effectively improve the signal-to-noise ratio of small moving targets and suppress background clutter and strong clutter under low signal-to-noise ratio conditions. It is particularly suitable for detecting small moving targets under low signal-to-noise ratio conditions.

[0017] The present invention is further described below in conjunction with specific implementation modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The figure is a flow chart of a small target detection method based on multiple hypothetical velocities of a gyroscope according to the present invention.

[0019] Figure 2 This is a flowchart of the FPGA hardware implementation of the solution of the present invention.

[0020] Figure 3 It is a schematic diagram of the implementation effect of each step in the embodiment of the present invention.

[0021] Figure 4 Schematic diagram of convolution operation implemented by FPGA hardware of the solution of the present invention.

[0022] Figure 5 is an original image containing noise in an embodiment of the present invention.

[0023] Figure 6 It is an accumulation diagram after assuming that the speed is accumulated in the embodiment of the present invention.

[0024] Figure 7 This is a diagram showing the effect of small target detection in an embodiment of the present invention, where Figure 7 (a) is the grayscale three-dimensional corresponding image after the target detection algorithm. Figure 7 (b) is the detection result of line correlation contrast based on centroid extraction. DETAILED DESCRIPTION

[0025] Example

[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an kind" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0028] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to those of ordinary skill in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be considered as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so that once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0029] Combination Figures 1 to 3 , a small target detection method based on multiple hypothesized velocities of a gyroscope, comprising the following steps:

[0030] Step 1: Calculate the optical axis model based on the multi-hypothesis speed of the gyroscope three-axis information, correspond the continuous images on the time axis to a coordinate system, perform multi-speed hypothesis accumulation on the image sequence under the same coordinate axis, slow down the target and improve the target signal-to-noise ratio:

[0031] Obtain gyroscope information, use the three-axis information data of the gyroscope to calculate the optical axis of the image, accurately calculate the camera shake compensation image motion, and thus restore a series of jittered image sequences on the time axis to the same coordinate system, so that the background clutter is stationary in the image and only the target itself moves;

[0032] It is found that the error does not change much in a very short time, so we consider using a linear model to correct the error. At this time, we consider that the target is also moving slowly. Since it is a long-distance small target imaging, we can describe the target movement as a slow uniform linear model. Therefore, for two linear uniform motion vectors, we add them together to merge them into a common linear uniform motion vector. In the FPGA implementation, the gyroscope information is received through the serial port, and the gyroscope three-axis information of each frame of the image is obtained by unpacking. The three-axis information is used to restore the jittered image sequence to the same coordinate system;

[0033] The three-axis rotation of the gyroscope is used to establish a three-axis matrix. In order to represent the three-axis rotation of an object, we can simply multiply the three single-axis rotation matrices.

[0034]

[0035] Among them, γ, θ, The rotation angles in the three directions of x, y, and z can be directly decoded by the gyroscope;

[0036] In FPGA, gyroscope information is received through the serial port, and the gyroscope three-axis information of each frame image is obtained by unpacking. The image coordinate system is established with the first frame, and the image is convolved from the second frame to correct the optical axis deviation caused by camera shake. Convolution module is established in FPGA, such as Figure 4 The figure shows a schematic diagram of convolution operation in FPGA. For the above 3*3 convolution, two shift ram IP cores are used to cache two lines of the input image. When the third line of image is output, it forms three lines of parallel image information together with the output of the two shift ram IP cores, and convolution operations are performed in sequence to restore the image sequence to the same coordinate system.

[0037] Figure 5 As shown in the figure, the small target in the original target image is submerged in the background and noise and is almost indistinguishable.

[0038] For the moving target in the image sequence under the same coordinate system, let the target speed be Vt, and assume multiple initial speeds Vg for it. After accumulation, the speed of the target in the image sequence is the combined effect of the target speed and the assumed initial speed. Because we use linear motion models to represent both Vg and Vt, we can use vector superposition to calculate the combined effect of the two speeds:

[0039] V=Vg+Vt

[0040] In order to accurately calculate the size and direction of the motion vector, we use multiple hypothetical speeds to approximate the speed. Generally, we set the speed size to the following 9 speeds, expressed in the form of (x, y), where x is the speed in the horizontal direction of the image, y is the speed in the vertical direction of the image, and the speed unit is pixel / frame. Take (0.1, 0.1) as an example to indicate that the hypothetical speed frame moves 0.1 pixel in the horizontal and vertical directions respectively:

[0041] (-0.1，-0.1) (-0.1，0) (-0.1，0.1) (0.1，-0.1) (0，0) (0，0.1) (0.1，-0.1) (0.1，0) (0.1，0.1)

[0042] Finally, the image sequence in the unified coordinate system is accumulated pixel by pixel to obtain multiple groups of accumulated images with different assumed initial velocities;

[0043] For FPGA, the speed corresponds to the direction and size of each frame displacement on the image. Therefore, an image displacement module can be designed as a multi-speed assumption module. For an image, we can divide the image displacement into rightward and downward displacement and leftward and upward displacement. When the image is displaced to the right, it can be regarded as the image being delayed and output. When the speed in the right direction is 1, it can be regarded as the image being delayed for one clock before being output. When the speed in the downward direction is 1, it can be regarded as the image being delayed for one line before being output. Another is displacement to the upper left. We can first delay the entire image backward, generate an advance signal from the undelayed image as the image output enable, and displace the delayed image to the upper left.

[0044] In order to realize the simultaneous accumulation of displacement in 9 directions, the FPGA designs an 18-row DDR read-write arbitration module in this embodiment, uses the MIG IP core to control the DDR, and forms the arbitration signal with the write request and read request of 18 data channels. When WR_REQ is detected to be high, the arbitration module outputs a write enable signal, and the DDR performs a write operation. When RD_REQ is detected to be high, the arbitration module outputs a read enable signal, and the DDR performs a read operation. This DDR accumulation adopts a partition design. Taking the initial velocity assumption that the x-axis direction is 1 and the y-axis direction is 1 as an example, the DDR is divided into two areas. One is an image area of ​​100 frames for storing a sequence of 100 images, and the 1-frame image position is used to store the accumulation map. In order to save the time of hardware calculation, after the accumulation of 1-100 frames is completed, when calculating the accumulation of 2-101 frames, the part of 2-100 frames has been accumulated for the first time. We can take the accumulation of 1-100 frames, add the 101st frame minus the 1st frame. This operation greatly reduces the amount of calculation of the FPGA and is conducive to meeting the real-time requirements of hardware implementation.

[0045] Figure 6 This is the cumulative graph after the target is effectively accumulated. The target capability is effectively accumulated and a weak tail line can be slightly seen.

[0046] Step 2: Perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target tail line information:

[0047] Step 2-1: The image accumulated in step 1 is easily disturbed by background information, so the accumulated image at intervals of m frames is extracted for image difference processing to obtain the initial tail line area in the difference image, which effectively reduces the interference of the background;

[0048] Since the target is moving and after the accumulated difference, the target will leave a light and dark line on the difference map, which is the initial trailing line obtained;

[0049] Step 2-2

[0050] The grayscale values ​​of points inside the line target have a certain correlation, while the grayscale values ​​of the line target and the surrounding background area have a large difference, which can be expressed by the grayscale gradient. The grayscale difference between the line target and the surrounding background is large, the three-dimensional surface gradient is large, and the correlation between the pixels in the background area is strong, the grayscale gradient between the pixels is small, and no mutation occurs;

[0051] For each point in the initial trailing line area, the difference between the grayscale value of the connected domain in different directions of 0°, 45°, 90°, and 135° centered on the point and the grayscale value of the surrounding area is determined:

[0052]

[0053] Among them, f(x, y) is the original image, g(θ) is the filter template of 0°, 45°, 90°, and 135° respectively, and g(x, y, θ) is the corresponding image after filtering;

[0054] Take 0° as an example to open the convolution window:

[0055]

[0056] The remaining direction line templates are obtained by rotating the 0° direction template.

[0057] Step 2-3, fuse the filtering results of each point in the four directions of 0°, 45°, 90°, and 135°, take the maximum value after filtering in the four directions as the gradient information of the pixel, and perform binary segmentation on the obtained gradient image:

[0058]

[0059] Where D(x,y) represents the image after four-directional filtering, and Th represents the binarization threshold;

[0060] For each pixel after four-directional filtering, a threshold comparison is performed, pixels greater than the threshold are set to 1, pixels less than the threshold are set to 0, and pixels that are not 0 are saved, that is, all target tail lines and suspected targets are processed together for subsequent connected domain processing.

[0061] Step 3: Extract the line target, calculate the length and slope of the trailing line, and use the line information to calculate the target's true speed:

[0062] Step 3-1, perform connected domain processing on the points with a value of 1 in the binary segmented image, and store the targets under the same connected domain, where the length of the connected domain is the length of the target's tail line, and the slope of the two end points of the connected domain is the target's movement direction;

[0063] In FPGA, the connected domain processing is implemented in a growing way. A 15*15 convolution kernel is opened and grows outward from the center point. When the detected line information is continuous and meets a certain length, we can consider the trailing line as a line target under the same connected domain, and record the starting coordinates and length information of the trailing line for storage. In this way, the connected domain algorithm can be implemented by convolution. This implementation method greatly shortens the FPGA operation time and improves the running speed of the algorithm.

[0064] Step 3-2: The signal-to-noise ratio of a small target is low. The energy of a moving small target is directly superimposed and dispersed. When the assumed speed is close to the target's true speed, its energy can be effectively gathered.

[0065] Compare the connected domain information under different speed assumptions. The group with the largest connected domain energy is the group whose assumed speed is closest to the target speed, and calculate the real speed of the target based on the trailing line information.

[0066] Assume that the angle between the trailing line and the positive x direction is α, and the angle between the velocity and x is β. The length component of the trailing line in the horizontal direction is L. x , the target's true horizontal velocity is V xt , the target's assumed horizontal velocity is V xh , when the accumulated number of frames is M, we have:

[0067]

[0068] L x =L·cosα#

[0069] L x =(V xt -V xh )·M#

[0070]

[0071] The length component of the tail line in the vertical direction is L y , the target's true vertical velocity is V yt ,but:

[0072] L y =L·sinα

[0073]

[0074] Among them, V xt Indicates the true speed of the target in the horizontal direction, V yt Indicates the true speed of the target in the vertical direction.

[0075] Step 4: Use the target's real motion information to perform reverse superposition, infer the target position, and suppress strong clutter:

[0076] Based on the actual speed of the target determined in step 3, the actual position of the target is determined by reverse calculation, and then the calculated actual speed of the target is used to replace the previously assumed initial speed of the target, and multiple frames of images are accumulated, and the point target area is obtained in the accumulated image;

[0077] For a line target, the target may have two moving directions. By reversely accumulating the two directions, the correct moving direction of the point target can be obtained.

[0078] Step 5: Use the line correlation contrast algorithm based on centroid extraction to detect small targets, suppress the background and improve the target contrast among many backgrounds and clutter:

[0079] Step 5-1. All target enhancement signal-to-noise ratio and denoising algorithms must eventually perform target detection. According to the target speed inferred above, the accumulated image is reversed to restore the point target according to the target speed in order to find the termination condition for reverse calculation. This paper adopts a patch contrast target detection algorithm to detect point targets after energy accumulation. In the process of restoring the target according to the target motion trajectory, we have basically determined the range of the target, and we perform point target detection on the area that may contain the target within a certain range;

[0080] This paper adopts the target neighborhood contrast detection algorithm. Since the international regulations stipulate that the size of small targets is 2*2-9*9, we open a 9*9 window and slide it in the potential target area. The 9*9 window is divided into the central area where the target exists and eight neighborhoods in eight directions of the central area. The neighborhood mean is calculated.

[0081] This paper focuses on the long-distance imaging of small targets. The size of the target on the camera target surface is only a few pixels. Therefore, a certain size area in the point target area obtained in step 4, that is, a 3*3 size area, is used as the center area of ​​the target, and the 8 3*3 neighborhoods of the same size around it are defined as the target background area. The average gray value of the center area is:

[0082]

[0083] Among them, m T represents the average gray value in the target area, N represents the number of pixels in the central area, F represents the gray value at (i, j), and T represents the target area;

[0084] The average gray value of the target background area is:

[0085]

[0086] Among them, m Bk Represents the average gray value of the target background area, N B Indicates the number of pixels in the background area;

[0087] Calculate the average grayscale value of the background area and the target area respectively, and determine the intensity difference between the background area and the target area based on this:

[0088] Z(x,y)=minl i

[0089] l1=d1·d9

[0090] l2=d2·d8

[0091] l3=d3·d7

[0092] l4=d4·d5

[0093]

[0094] The above operation is implemented in FPGA by opening a 9*9 convolution window, calculating the average grayscale value of the background area and the target area respectively, calculating the line contrast of the point target after subtraction, and replacing the value of the central pixel with the line contrast value calculated in the convolution window to enter the centroid detection operation.

[0095] Step 5-2: Determine the target centroid based on the intensity difference between the background area and the target area. Considering the real-time nature of the algorithm, this paper uses a sliding window to traverse the image instead of the 8-connected domain centroid detection algorithm. Use a fixed window to traverse the image, compare the intensity value of the central pixel with the intensity value of its neighboring pixels, and determine whether it is the maximum value of the current window. If so, keep the pixel, otherwise remove it:

[0096]

[0097] Among them, U is the convolution area, maxZ(i,j) is the maximum contrast intensity between each pixel in the convolution area and its surroundings;

[0098] Step 5-3: Determine the small target in the image based on the determined target centroid and the image after binary segmentation:

[0099] CLPCM(x,y)=D(x,y)×C(x,y)

[0100]

[0101] Among them, T is the local area of ​​the target trajectory, and Th is the set threshold.

[0102] Figure 7 (a) is the grayscale three-dimensional response map of the image after centroid extraction. The target is effectively enhanced and the background is weakened. Figure 7 (b) The target is screened out after binarization, and the background and noise are suppressed.

[0103] The present invention also provides a small target detection system based on multiple hypothetical velocities of a gyroscope, comprising the following steps:

[0104] Multi-speed accumulation module: used to calculate the optical axis model based on the multi-hypothesis speed of the gyroscope's three-axis information, correspond the continuous images on the time axis to a coordinate system, perform multi-speed hypothesis accumulation on the image sequence under the same coordinate axis, slow down the target and improve the target signal-to-noise ratio;

[0105] Trailing line extraction module: used to perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target trailing line information;

[0106] True speed calculation module: used to extract line targets, calculate the length and slope of the trailing line, and calculate the true speed of the target using line information;

[0107] Target position determination module: used to utilize the real motion information of the target to perform reverse superposition, infer the target position, and suppress strong clutter;

[0108] Target detection module: It is used to detect weak targets by using line correlation contrast detection based on centroid extraction, suppressing the background and improving the target contrast among many backgrounds and clutter.

[0109] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0110] Step 1: Based on the gyroscope three-axis information, the optical axis model is calculated by multi-hypothesis speed, the continuous images on the time axis correspond to a coordinate system, the image sequence under the same coordinate axis is accumulated by multi-speed hypothesis, and the target is decelerated to improve the target signal-to-noise ratio;

[0111] Step 2: Perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target trailing line information;

[0112] Step 3: extract the line target, calculate the length and slope of the trailing line, and use the line information to calculate the target's true speed;

[0113] Step 4: Use the real motion information of the target to perform reverse superposition, infer the target position, and suppress strong clutter;

[0114] Step 5: Use the line correlation contrast algorithm based on centroid extraction to detect small targets, suppress the background and improve the target contrast among many backgrounds and clutter.

[0115] The present invention also provides a computer storable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0116] Step 1: Based on the gyroscope three-axis information, the optical axis model is calculated by multi-hypothesis speed, the continuous images on the time axis correspond to a coordinate system, the image sequence under the same coordinate axis is accumulated by multi-speed hypothesis, and the target is decelerated to improve the target signal-to-noise ratio;

[0117] Step 2: Perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target trailing line information;

[0118] Step 3: extract the line target, calculate the length and slope of the trailing line, and use the line information to calculate the target's true speed;

[0119] Step 4: Use the real motion information of the target to perform reverse superposition, infer the target position, and suppress strong clutter;

[0120] Step 5: Use the line correlation contrast algorithm based on centroid extraction to detect small targets, suppress the background and improve the target contrast among many backgrounds and clutter.

[0121] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A small target detection method based on multi-hypothesis velocity of gyroscope, characterized in that: The following steps are involved: Step 1: Based on the gyroscope three-axis information, the optical axis model is calculated by multi-hypothesis speed, the continuous images on the time axis correspond to a coordinate system, the image sequence under the same coordinate axis is accumulated by multi-speed hypothesis, and the target is decelerated to improve the target signal-to-noise ratio; Step 2: Perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target trailing line information; Step 3: extract the line target, calculate the length and slope of the trailing line, and use the line information to calculate the target's true speed; Step 4: Use the real motion information of the target to perform reverse superposition, infer the target position, and suppress strong clutter; Step 5: Use the line correlation contrast algorithm based on centroid extraction to detect small targets, suppress the background and improve the target contrast among many backgrounds and clutter.

2. The method for detecting small and weak targets based on multiple hypothetical velocities of a gyroscope according to claim 1, characterized in that: The optical axis correction multi-speed accumulation in step 1 is specifically as follows: Obtain gyroscope information, use the three-axis information data of the gyroscope to calculate the optical axis of the image, and thus restore a series of jittered image sequences on the time axis to the same coordinate system; For the moving target in the image sequence under the same coordinate system, let the target speed be Vt, and assume multiple initial speeds Vg for it. After accumulation, the speed of the target in the image sequence is the combined effect of the target speed and the assumed initial speed: V=Vg+Vt Finally, the image sequence in the unified coordinate system is accumulated pixel by pixel to obtain multiple groups of accumulated images with different assumed initial velocities.

3. The method for detecting small and weak targets based on multiple hypothesized velocities of a gyroscope according to claim 1, characterized in that: The step 2 of obtaining the target trailing line information is as follows: Step 2-1, for the image accumulated in step 1, extract the accumulated image at intervals of m frames and perform image difference processing to obtain the initial tail line area in the difference image; Step 2-2: For each point in the initial trailing line area, determine the difference between the grayscale value of the connected domain in different directions of 0°, 45°, 90°, and 135° with the point as the center and the grayscale value of the surrounding area: Where f(x, y) is the original image, g(θ) is the filter template of 0°, 45°, 90°, and 135° respectively, and g(x, y, θ) is the corresponding image after filtering; Step 2-3, fuse the filtering results of each point in the four directions of 0°, 45°, 90°, and 135°, take the maximum value after filtering in the four directions as the gradient information of the pixel, and perform binary segmentation on the obtained gradient image: Where D(x, y) represents the image after four-directional filtering, and Th represents the binarization threshold; For each pixel after four-directional filtering, a threshold comparison is performed, pixels greater than the threshold are set to 1, pixels less than the threshold are set to 0, and pixels that are not 0 are saved, that is, all target tail lines and suspected targets are processed together for subsequent connected domain processing.

4. The method for detecting small and weak targets based on multiple hypothetical velocities of a gyroscope according to claim 3, characterized in that: The calculation of the target true speed using line information in step 3 is specifically as follows: Step 3-1, perform connected domain processing on the points with a value of 1 in the binary segmented image, and store the targets under the same connected domain, where the length of the connected domain is the length of the target's tail line, and the slope of the two end points of the connected domain is the target's movement direction; Step 3-2, compare the connected domain information under different speed assumptions, the group with the largest connected domain energy is the group whose assumed speed is closest to the target speed, and calculate the real speed of the target based on the trailing line information; Assume that the angle between the trailing line and the positive x direction is α, and the angle between the velocity and x is β. The length component of the trailing line in the horizontal direction is L. x , the target's true horizontal velocity is V xt , the target's assumed horizontal velocity is V xh , when the accumulated number of frames is M, we have: L x =L·cosα# L x =(V xt -V xh )·M# The length component of the tail line in the vertical direction is L y , the target's true vertical velocity is V yt ,but: L y =L·sinα Among them, V xt Indicates the true speed of the target in the horizontal direction, V yt Indicates the true speed of the target in the vertical direction.

5. The method for detecting small and weak targets based on multiple hypothetical velocities of a gyroscope according to claim 3, characterized in that: The step 4 uses the real motion information of the target to perform reverse superposition and infer the target position, specifically: Based on the actual speed of the target determined in step 3, the actual position of the target is determined by inverse calculation. Then, the calculated actual speed of the target replaces the assumed initial speed of the target, multiple frames of images are accumulated, and the point target area is obtained in the accumulated image.

6. The method for detecting small and weak targets based on multiple hypothetical velocities of a gyroscope according to claim 5, characterized in that: The small target detection in step 5 is specifically as follows: Step 5-1: Take a certain size area in the point target area obtained in step 4 as the central area of ​​the target, and define the eight surrounding areas of the same size as the target background area. The average gray value of the central area is: Among them, m r represents the average gray value in the target area, N represents the number of pixels in the central area, F represents the gray value at (i, j), and T represents the target area; The average gray value of the target background area is: Among them, m Bk Represents the average gray value of the target background area, N B Indicates the number of pixels in the background area; Calculate the average grayscale value of the background area and the target area respectively, and determine the intensity difference between the background area and the target area based on this: Z(x,y) min l i l1=d1·d9 l2=d2·d8 l3=d3·d7 l4=d4·d5 Step 5-2: Determine the target centroid based on the intensity difference between the background area and the target area: Among them, U is the convolution area, maxZ(i, j) is the maximum contrast intensity between each pixel in the convolution area and its surroundings; Step 5-3: Determine the small target in the image based on the determined target centroid and the image after binary segmentation: CLPCM(x,y)=D(x,y)×C(x,y) Among them, T is the local area of ​​the target trajectory, and Th is the set threshold.

7. A small target detection system based on gyroscope multi-hypothesis velocity, characterized in that: The following steps are involved: Multi-speed accumulation module: used to calculate the optical axis model based on the multi-hypothesis speed of the gyroscope's three-axis information, correspond the continuous images on the time axis to a coordinate system, perform multi-speed hypothesis accumulation on the image sequence under the same coordinate axis, slow down the target and improve the target signal-to-noise ratio; Trailing line extraction module: used to perform differential processing on the accumulated image, eliminate background information on the differential image, detect line targets, and obtain target trailing line information; True speed calculation module: used to extract line targets, calculate the length and slope of the trailing line, and calculate the true speed of the target using line information; Target position determination module: used to utilize the real motion information of the target to perform reverse superposition, infer the target position, and suppress strong clutter; Target detection module: It is used to detect weak targets by using line correlation contrast detection based on centroid extraction, suppressing the background and improving the target contrast among many backgrounds and clutter.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer storable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images