A method and system for detecting the trajectory of weak moving targets based on sonar images

Abstract

The present invention provides a method and system for detecting the trajectory of a weak moving target based on sonar images. The method includes: using the shape parameter, scale parameter in the gray level and a preset false alarm rate as the input parameters of the CA-CFAR detector; taking a rectangular window as the selection criterion for the reference unit of the target detection sonar image, traversing each pixel value on the target detection sonar image to obtain the accumulated value of the echo intensity; when the accumulated value of the echo intensity is greater than or equal to the preset detection threshold, discarding the trajectory with the minimum energy, and splicing the echo intensities on the remaining trajectories; when the accumulated value of the echo intensity is less than the preset detection threshold, splicing all the trajectories to obtain the corresponding reference unit; estimating the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtaining the estimated value of the detection threshold; comparing the detection unit with the estimated value of the detection threshold to obtain the detection result.

Description

Technical Field

[0001] The present invention belongs to the technical field of sonar image processing, and particularly relates to a method and system for detecting the trajectory of a weak moving target based on sonar images. Background Art

[0002] The detection and tracking of underwater moving targets are hot topics in the current field of target detection. It has great significance and wide applications in both civilian and military fields, such as the detection of frogmen in the offshore environment, the detection of the movement trajectories of fish schools, and the early warning of underwater moving targets such as UUVs (Unmanned Underwater Vehicles). Due to the low underwater visibility, the optical imaging effect is poor. Therefore, acoustic wave devices are usually adopted for detection in the current underwater target detection field. Among them, the forward-looking sonar is the main current underwater moving small target detection device. However, affected by the complex underwater environment, such as complex seabed reverberation, scattering of sound propagation, and acoustic wave attenuation, especially the influence of seabed reverberation, the resolution of sonar images is low and the signal-to-noise ratio of the detected underwater targets is weak. Therefore, there are still great difficulties in the current detection and tracking of underwater weak targets.

[0003] Traditional moving target detection and tracking technologies are all based on the strategy of detection before tracking (DBT). This technology performs well and has a simple algorithm in the case of relatively high signal-to-noise ratio. However, affected by the weak target intensity and complex environment in sonar images, there are a large number of false speckle noises in the images. If traditional detection and tracking technologies are used to perform threshold detection on each frame, a large number of missed alarms or false alarms will occur. For this reason, in the current radar weak target detection, the technology of tracking before detection is widely used and can better handle the detection problem of weak targets. This method estimates the adaptive threshold for each detection unit after obtaining multiple frames of data, and then performs threshold detection on all point tracks. While detecting the trajectory, the weak moving target is also detected. In the weak target detection of radar, the detection can be carried out according to the high-speed movement characteristics of the target and by using Doppler information. However, the movement speed of underwater targets is very low, the echo is weak, and the background is complex, so it is impossible to use Doppler information, and thus the detection is very difficult.

[0004] Therefore, in recent years, more and more researchers have imitated the radar weak target detection technology to study the detection technology of small and weak targets in sonar images. Cui Jie (published in "Moving Target Detection of Multibeam Forward-Looking Sonar Based on Frame Difference Method") proposed a background elimination algorithm based on frame difference method for the problem of moving target detection of multibeam forward-looking sonar. It has good performance when the sonar device is stationary, but this method has great limitations. When the environment changes or the target echo is weak, there will be a high false alarm rate. Jing (published in "A method to estimate the abundance of fish based on dual-frequency identification sonar (DIDSON) imaging") combined the nearest neighbor algorithm and extended Kalman filter to track multiple fish school targets, which has high accuracy and certain convenience in the detection of fish schools, and designed an algorithm for data association to track multiple targets in 3D space (published in "A Method to Track Targets in Three-Dimensional Space Using an Imaging Sonar"), showing good performance when there are fewer targets. However, the data association-based method is prone to the situation where multiple targets and sonar beams are on the same spherical surface, resulting in overlapping target areas in the sonar image. Therefore, it is difficult to track multiple dense underwater targets. Zhou Tian et al. (published in "Gaussian Filtering Algorithm for Multi-Target Tracking in Sonar Images") proposed an improved Gaussian mixture probability hypothesis density filtering algorithm for the problem of "unknown intensity of newly born targets" in multi-target tracking of sonar images. It adopts the strategy of measurement-driven new birth and has the performance of lower error and faster operation. However, this method requires good measurement as a prerequisite. When dealing with weak targets, the measurement effect is not good, and the inability to correctly segment them will greatly affect the generation of new targets.

[0005] According to the above description, the target echo intensity in sonar images largely determines the lower limit of detection. Facing this situation, the CFAR (Constant False Alarm Rate) detector, as a type of TBD (Track Before Detect) algorithm, can utilize the information of background reference cells to reduce the difficulty of detection. ROBEY (published in "A CFAR Adaptive Matched Filter Detector") first proposed the CFAR adaptive detector and applied it to radar target detection. Subsequently, Sebastián A. Villar (published in "Pipeline detection system from acoustic images utilizing CA-CFAR") applied the CA-CFAR (Cell Averaging Constant False Alarm Rate) detection technology to the pipeline target detection of sidescan sonar images and proposed a new reference cell selection strategy, but it has a certain amount of computation. For this reason, Acosta (published in "Accumulated CA–CFAR Process in 2-D for Online Object Detection From Sidescan Sonar Data") applied the CFAR detection technology to the pipeline detection of sidescan sonar (SSS) images and proposed a 2-D (two-dimensional) cumulative cell averaging constant false alarm detection algorithm (ACA-CFAR 2-D), which optimized the computation of CA-CFAR and had good performance in high-resolution sidescan sonar images. Sebastián A. Villar (published in "OS-CFAR Process in 2-D for Object Segmentation from Sidescan Sonar Data") also applied the OS-CFAR technology to the target segmentation of sonar images and had good performance on high-resolution sidescan images, but was restricted by the image signal-to-noise ratio. Although OS-CFAR has a good effect on suppressing speckle noise, it sacrifices computational efficiency for this. Therefore, the author accelerated the sample sorting problem by improving the calculation of the statistical order (published in "Efficient Approach for OS-CFAR 2D Technique Using Distributive Histograms and Breakdown Point Optimal Concept applied to Acoustic Images"), improving the operation efficiency.The objects of the above researchers are all to detect large targets in the high-resolution images of sidescan sonar. For this reason, Zheng (published in "Detection of Small Objects in Sidescan Sonar Images Based on POHMT and Tsallis Entropy") further studies small targets in high-resolution sidescan images. Morphological POHMT is used to enhance weak small objects, and Tsallis entropy considering the long-range interaction between particles in the system is used to determine the optimal threshold of POHMT, which has relatively good performance in detecting small targets. However, the algorithm performance is still restricted by the signal-to-noise ratio. Currently, the CFAR algorithm is widely used in the detection of sonar image targets, but it is basically limited by the quality of sonar images. In the case of strong target echoes, a smaller false alarm rate can be set, so clutter can be suppressed. However, if the target is submerged in reverberation, setting a larger false alarm rate may detect the target but will inevitably introduce a large number of false alarm targets. Therefore, it is urgent to study the target detection technology for sonar images with low quality and weak targets.

[0006] Generally speaking, the detection of small and weak moving targets underwater based on sonar images is an important topic in current underwater target detection. However, affected by the harsh underwater environment, especially seabed reverberation, the detection technology is very difficult. Therefore, an improved CFAR algorithm is urgently needed. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to propose a method and system for detecting the trajectory of weak moving targets based on sonar images, so as to improve the accuracy of identifying the movement trajectory of weak moving targets.

[0008] Based on the above purpose, the present invention proposes a method for detecting the trajectory of weak moving targets based on sonar images, including:

[0009] Based on the fact that the gray level of reverberation noise in the sonar image for target detection follows a Weibull distribution, the shape parameter, the scale parameter in the gray level, and a preset false alarm rate are used as the input parameters of the CA-CFAR detector;

[0010] Taking a rectangular window as the selection criterion for the reference unit of the sonar image for target detection, traversing each pixel value on the sonar image for target detection to obtain the cumulative value of echo intensity, and the rectangular window selects the reference unit in a sliding form;

[0011] When the accumulated echo intensity value is greater than or equal to the preset detection threshold, discard the trajectory with the minimum energy, splice the echo intensities on the remaining trajectories, and obtain the corresponding reference cell; when the accumulated echo intensity value is less than the preset detection threshold, splice all the trajectories to obtain the corresponding reference cell;

[0012] Estimate the reference cell through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtain the detection threshold estimation value;

[0013] Compare the detection cell with the detection threshold estimation value to obtain the detection result.

[0014] In some embodiments, based on the fact that the gray level of the reverberation noise of the target detection sonar image follows a Weibull distribution, the shape parameter, the scale parameter in the gray level, and the preset false alarm rate are used as the input parameters of the CA-CFAR detector, including:

[0015] For the probability density function of the Weibull distribution and the cumulative distribution function where x is the random variable, β is the shape parameter used to describe the aggregation degree of the reverberation data, η is the scale parameter used to describe the position point of the reverberation median, the probability density function of the mean statistic Y of the reference cell data is obtained as where Γ(θ) = ∫0 ∞ t θ-1 e -t dt, where θ is the function variable and Y is the mean statistic of the reference cell data;

[0016] Calculate the nominal factor through the false alarm rate equation in the CA-CFAR detector;

[0017] Through integral derivation, obtain from the nominal factor where η is the scale parameter, PFA is the false alarm rate, and β is the shape parameter.

[0018] In some embodiments, taking the rectangular window as the selection criterion for the reference cell of the target detection sonar image, traverse each pixel value on the target detection sonar image to obtain the accumulated echo intensity value, and the rectangular window selects the reference cell in a sliding form, including:

[0019] Taking the rectangular window as the selection criterion for the reference cell of the target detection sonar image, starting from the center point of the rectangular window, select four trajectories at four angles of the cross shape at intervals of 45 degrees to accumulate the echo intensity values;

[0020] Starting from the center point of the rectangular window, a range-azimuth two-dimensional reference window is formed, the echo intensity values of a predetermined number of trajectories are accumulated, and sorted from small to large. Through the formula

[0021] the cumulative echo intensity value is calculated, where i and j are the abscissa and ordinate of the pixel points of the target to be detected in the target detection sonar image respectively, n is the number of motion trajectories, image(i,j) represents the echo intensity value of the pixel point (i,j), and track1 to trackn represent the 1st to nth motion trajectories respectively.

[0022] In some embodiments, the reference unit is estimated by maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and the detection threshold estimation value is obtained, including:

[0023] Using the bisection method to solve the transcendental equation formed by the maximum likelihood function of the reference unit;

[0024] Iteratively calculating the shape parameter and scale parameter until reaching a steady state to obtain the detection threshold estimation value.

[0025] The present invention also proposes a method for detecting the trajectory of a weak moving target based on a sonar image, including:

[0026] Obtaining multiple frames of sonar images of a moving target, and performing a joint operation on the multiple frames of sonar images through the maximum selection and accumulation algorithm to obtain a detection image of the moving target;

[0027] Using the method as described above to perform target point detection on the detection image;

[0028] Determining the motion trajectory of the moving target through a trajectory detector.

[0029] In some embodiments, obtaining multiple frames of sonar images of a moving target, and performing a joint operation on the multiple frames of sonar images through the maximum selection and accumulation algorithm to obtain a detection image of the moving target, including:

[0030] Traversing multiple frames of sonar images of the moving target within a specified time range, analyzing the range-azimuth information, and constructing a range-azimuth-time three-dimensional information model;

[0031] Performing the maximum selection and accumulation algorithm on the range-azimuth-time three-dimensional information model to obtain a detection image of the moving target.

[0032] In some embodiments, determining the motion trajectory of the moving target through a trajectory detector, including:

[0033] Performing median filtering on the detection result to obtain a median filtering result;

[0034] Perform a Hough transform on the median filtering result to obtain a trajectory detection result.

[0035] Based on the above object, the present invention also provides a system for detecting the trajectory of a weak moving target based on a sonar image, including:

[0036] An input module, configured to use the shape parameter, the scale parameter, and a preset false alarm rate in the gray level as input parameters of a CA-CFAR detector, based on the fact that the gray level of the reverberation noise of the target detection sonar image follows a Weibull distribution;

[0037] A selection module, configured to use a rectangular window as a selection reference for the reference unit of the target detection sonar image, traverse each pixel value on the target detection sonar image, obtain an echo intensity accumulation value, and select the reference unit in a sliding form;

[0038] A judgment module, configured to discard the trajectory with the minimum energy and splice the echo intensities on the remaining trajectories to obtain a corresponding reference unit when the echo intensity accumulation value is greater than or equal to a preset detection threshold; and splice all the trajectories to obtain a corresponding reference unit when the echo intensity accumulation value is less than the preset detection threshold;

[0039] An estimation module, configured to estimate the reference unit through maximum likelihood estimation to obtain the shape parameter and the scale parameter of the reverberation noise distribution, and obtain a detection threshold estimation value;

[0040] A comparison module, configured to compare the detection unit with the detection threshold estimation value to obtain a detection result.

[0041] In some embodiments, the estimation module includes:

[0042] A solution unit, configured to solve the transcendental equation formed by the maximum likelihood function of the reference unit using the bisection method;

[0043] An iteration unit, configured to iteratively calculate the shape parameter and the scale parameter until reaching a steady state to obtain a detection threshold estimation value.

[0044] The present invention also provides a system for detecting the trajectory of a weak moving target based on a sonar image, characterized by including:

[0045] A maximum accumulation module, configured to obtain multiple frames of sonar images of a moving target, perform a joint operation on the multiple frames of sonar images through a maximum accumulation algorithm, and obtain a detection image of the moving target;

[0046] A detection module, configured to perform target point detection on the detection image using the method described above;

[0047] A result module, configured to determine a motion trajectory of the moving target through a trajectory detector.

[0048] Generally speaking, the beneficial effects of the present invention are as follows: Aiming at the problem of poor detection performance of small and weak targets in current sonar images, the present invention proposes a brand-new CFAR detection system, which applies the prior knowledge of the pixel distribution of seabed reverberation to regional noise estimation. First, the continuous-frame sonar images are combined by adopting a strategy of maximum accumulation, and a new method for selecting reference cells is proposed. The energy mean difference of different predicted trajectories is used to eliminate the possible motion trajectories of the target, preventing the influence of target trajectory pixels on detection. Secondly, an improved CFAR detector is proposed, which uses the distribution information of regional noise to perform adaptive threshold estimation on the cell to be detected. Finally, the detection points of the sonar image are obtained, median filtering is used for processing, and then the Hough transform is used for trajectory detection. The gray pixel distribution characteristics of the reverberation noise in the measured sonar image are verified through experiments, verifying the feasibility of the algorithm. In addition, through the comparison of the detection performance of each algorithm at the end, the improved reference cell selection strategy and the improved CFAR detector proposed by the present invention have better detection effects, can detect weaker targets, and have better reverberation suppression effects. Description of the Drawings

[0049] In the drawings, unless otherwise specified, the same reference numerals throughout the several views denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in accordance with the present invention and should not be regarded as limiting the scope of the present invention.

[0050] Figure 1 A flowchart showing a method for detecting the trajectory of a weak moving target based on sonar images according to an embodiment of the present invention.

[0051] Figure 2 A flowchart showing another method for detecting the trajectory of a weak moving target based on sonar images according to an embodiment of the present invention.

[0052] Figure 3 A block diagram showing the composition of a system for detecting the trajectory of a weak moving target based on sonar images according to an embodiment of the present invention.

[0053] Figure 4 A block diagram showing the composition of another system for detecting the trajectory of a weak moving target based on sonar images according to an embodiment of the present invention.

[0054] Figure 5 A block diagram showing the composition of an estimation module according to an embodiment of the present invention.

[0055] Figure 6 A structural diagram showing the composition of a classical CFAR detector.

[0056] Figure 7 Shows a schematic diagram of the selection of the distance-azimuth reference unit according to an embodiment of the present invention.

[0057] Figure 8 Shows a sonar sequence image moving target trajectory detection system according to an embodiment of the present invention.

[0058] Figure 9a Shows the maximum accumulation effect of the sonar image in the horizontal movement direction.

[0059] Figure 9b Shows the maximum accumulation effect of the sonar image in the vertical movement direction.

[0060] Figure 9c Shows the maximum accumulation effect of the sonar image in the oblique movement direction.

[0061] Figure 10a Shows the measured seabed reverberation image.

[0062] Figure 10b Shows the gray histogram fitting diagram of the reverberation.

[0063] Figure 11a Shows the visualization result diagram of a single-frame forward-looking sonar image.

[0064] Figure 11b Shows the three-dimensional visualization result diagram of the single-target echo intensity.

[0065] Figure 11c Shows the three-dimensional visualization result diagram of a single-frame forward-looking sonar image.

[0066] Figure 12a Shows the detection result diagram of the maximum accumulation sonar image of the ten-frame combined image with the target moving in the horizontal direction.

[0067] Figure 12b Shows the detection result diagram of the rectangular window - traditional CA-CFAR of the ten-frame combined image with the target moving in the horizontal direction.

[0068] Figure 12c Shows the detection result diagram of the cross window - traditional CA-CFAR of the ten-frame combined image with the target moving in the horizontal direction.

[0069] Figure 12d Shows the detection result diagram of the manually set threshold detection of the ten-frame combined image with the target moving in the horizontal direction.

[0070] Figure 12e Shows the detection result diagram of the ideal detection result of the ten-frame combined image with the target moving in the horizontal direction.

[0071] Figure 12fDetection result graph of rectangular window - Weibull - CA - CFAR for ten - frame combined images showing the target moving horizontally.

[0072] Figure 12g Detection result graph of global threshold estimation for ten - frame combined images showing the target moving horizontally.

[0073] Figure 12h Detection result graph showing the detection results of ten - frame combined images with the target moving horizontally according to the embodiments of the present invention.

[0074] Figure 13a Detection result graph of trajectory detection for the maximum - selection cumulative sonar image with the target moving horizontally.

[0075] Figure 13b Detection result graph of trajectory detection for rectangular window - traditional CA - CFAR with the target moving horizontally.

[0076] Figure 13c Detection result graph of trajectory detection for cross - window - traditional CA - CFAR with the target moving horizontally.

[0077] Figure 13d Detection result graph of trajectory detection for manually - set threshold detection with the target moving horizontally.

[0078] Figure 13e Detection result graph of trajectory detection for ideal detection results with the target moving horizontally.

[0079] Figure 13f Detection result graph of trajectory detection for rectangular window - Weibull - CA - CFAR with the target moving horizontally.

[0080] Figure 13g Detection result graph of trajectory detection for global threshold estimation with the target moving horizontally.

[0081] Figure 13h Detection result graph of trajectory detection with the target moving horizontally according to the embodiments of the present invention.

[0082] Figure 14a Detection result graph of the maximum - selection cumulative sonar image with the target moving vertically.

[0083] Figure 14b Detection result graph of rectangular window - traditional CA - CFAR with the target moving vertically.

[0084] Figure 14c Detection result graph of cross - window - traditional CA - CFAR with the target moving vertically.

[0085] Figure 14dShows the detection result graph of the artificially set threshold detection where the target moves in the vertical direction.

[0086] Figure 14e Shows the ideal detection result graph where the target moves in the vertical direction.

[0087] Figure 14f Shows the detection result graph of the rectangular window - Weibull - CA - CFAR where the target moves in the vertical direction.

[0088] Figure 14g Shows the detection result graph of the global threshold estimation where the target moves in the vertical direction.

[0089] Figure 14h Shows the detection result graph where the target moves in the vertical direction according to the embodiment of the present invention.

[0090] Figure 15a Shows the trajectory detection result graph of the maximum - accumulation sonar image where the target moves in the vertical direction.

[0091] Figure 15b Shows the trajectory detection result graph of the rectangular window - traditional CA - CFAR where the target moves in the vertical direction.

[0092] Figure 15c Shows the trajectory detection result graph of the cross - window - traditional CA - CFAR where the target moves in the vertical direction.

[0093] Figure 15d Shows the trajectory detection result graph of the artificially set threshold detection where the target moves in the vertical direction.

[0094] Figure 15e Shows the trajectory detection result graph of the ideal detection result where the target moves in the vertical direction.

[0095] Figure 15f Shows the trajectory detection result graph of the rectangular window - Weibull - CA - CFAR where the target moves in the vertical direction.

[0096] Figure 15g Shows the trajectory detection result graph of the global threshold estimation where the target moves in the vertical direction.

[0097] Figure 15h Shows the trajectory detection result graph where the target moves in the vertical direction according to the embodiment of the present invention.

[0098] Figure 16a Shows the detection result graph of the maximum - accumulation sonar image where the target moves in the oblique direction.

[0099] Figure 16b Shows the detection result graph of the rectangular window - traditional CA - CFAR where the target moves in the oblique direction.

[0100] Figure 16c Cross-window showing the target moving in an oblique direction - Detection result graph of traditional CA-CFAR.

[0101] Figure 16d Detection result graph of manually set threshold detection showing the target moving in an oblique direction.

[0102] Figure 16e Ideal detection result graph showing the target moving in an oblique direction.

[0103] Figure 16f Rectangle window - Detection result graph of Weibull-CA-CFAR showing the target moving in an oblique direction.

[0104] Figure 16g Detection result graph of global threshold estimation showing the target moving in an oblique direction.

[0105] Figure 16h Detection result graph showing the target moving in an oblique direction according to an embodiment of the present invention.

[0106] Figure 17a Detection result graph of trajectory detection of the maximum cumulative sonar image showing the target moving in an oblique direction.

[0107] Figure 17b Rectangle window - Detection result graph of trajectory detection of traditional CA-CFAR showing the target moving in an oblique direction.

[0108] Figure 17c Cross-window - Detection result graph of trajectory detection of traditional CA-CFAR showing the target moving in an oblique direction.

[0109] Figure 17d Detection result graph of trajectory detection of manually set threshold detection showing the target moving in an oblique direction.

[0110] Figure 17e Detection result graph of trajectory detection of ideal detection result showing the target moving in an oblique direction.

[0111] Figure 17f Rectangle window - Detection result graph of trajectory detection of Weibull-CA-CFAR showing the target moving in an oblique direction.

[0112] Figure 17g Detection result graph of trajectory detection of global threshold estimation showing the target moving in an oblique direction.

[0113] Figure 17h Detection result graph of trajectory detection showing the target moving in an oblique direction according to an embodiment of the present invention.

[0114] Figure 18Shows a comparison chart of the detection effects between the embodiments of the present invention and traditional detection algorithms. Detailed implementation manners

[0115] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. Additionally, it should be noted that for ease of description, only parts related to the relevant invention are shown in the drawings.

[0116] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and embodiments.

[0117] In a sonar imaging system, a sonar device detects by emitting sound waves within a certain distance - azimuth angle range, and then switches from the emission mode to the reception mode to receive the echo intensity in the distance - azimuth to visualize the situation within the detection range. Observation units representing distance - azimuth are presented on the sonar image, and each observation unit represents the target echo intensity at a certain actual distance - azimuth. Subsequently, the detection system determines the presence or absence of a target in each observation unit through a pre - set detection threshold. In most detection systems, the principle of hypothesis testing is usually used to judge each observation unit. As shown in Equation (1), assume that event H1 represents the presence of a target in the detection unit, and H0 represents the absence of a target.

[0118] H1:X i >T

[0119] H0:X i ≤T (1)

[0120] Where Xi represents the intensity value of the observation unit, and T represents the detection threshold set by the detection system. This strategy is a traditional detection method. However, affected by the attenuation of sound wave propagation and severe seabed reverberation, the target intensity in the sonar image is weak and the reverberation interference is serious, resulting in high false alarm rates and missed alarm rates often occurring in the detection of weak targets underwater.

[0121] The current CFAR detection technology can better solve this problem. The CFAR detection technology estimates the noise through reference units near the target observation unit, and then obtains the adaptive detection threshold for each observation unit. This strategy can obviously show excellent performance in the complex and changeable underwater environment. Additionally, since the underwater movement speed of the target is relatively slow, within a short period of time, the targets detected by the sonar move in a straight line. Therefore, the present invention performs trajectory detection on the detected point traces based on the Hough transform.

[0122] The classic Constant False Alarm Rate (CFAR) detection algorithm performs detection on a single-frame signal. The threshold value is estimated through reference cells, and then the detection cell is compared and tested. Its working principle is as Figure 6 shown. Where D represents the cell to be detected, P represents the protection cell, N represents the length of the sliding window, X = [x1, x2,..., x N represents the data of N reference cells, Y is the background power reference value obtained from the statistics of N reference cells, α is the nominal factor, and T is the regional detection threshold obtained by CFAR. In traditional CFAR, the setting of the background power reference value is different for different algorithms. In CA-CFAR, the nominal factor is controlled by the clutter distribution. For example, in a classic radar detection system, it is known that the amplitude values of clutter data follow an exponential distribution, and its probability density function and cumulative distribution function are as follows:

[0123]

[0124] And from it can be seen that the observation cells are independent and identically distributed. According to the convolution formula of multiple random variables, the probability density function of the statistic of the reference cells can be obtained as:

[0125]

[0126] Therefore, after setting a constant false alarm rate (PFA), the nominal factor can be obtained from the false alarm rate equation calculated by CA-CFAR:

[0127]

[0128] From the joint distribution of equations (2) and (3), it can be obtained:

[0129]

[0130] Finally, by performing integration by parts on equation (5) above, the relationship between the false alarm rate and the nominal factor is obtained as follows:

[0131]

[0132] Therefore, after presetting the false alarm rate, the nominal factor can be solved through equation (6) Finally, the observed value D of the comparison detection cell and the adaptive detection threshold T are compared by a comparator, and then it is determined whether the target exists.

[0133] It is worth mentioning that, in addition to CA-CFAR, other traditional CFAR detection algorithms, such as OS-CFAR (Ordered Statistics Constant False Alarm Rate), GO-CFAR (Greatest of Means Constant False Alarm Rate), and ACA-CFAR (Accumulative Cell Averaging Constant False Alarm Rate), etc., by setting different statistical variables of the reference cells for the background noise in different scenarios, combining the probability density function distribution of the clutter amplitude, obtain the corresponding nominal factor and the adaptive detection threshold through the above-mentioned principle of solving the adaptive detection threshold. It can be seen from the current performance of the detection system that the CFAR detection algorithm shows excellent performance in dealing with the situation where the target echo intensity is weak, which is convenient for implementation in engineering applications.

[0134] Figure 1 FIG. shows a flowchart of a method for detecting the trajectory of a weak moving target based on a sonar image according to an embodiment of the present invention. As Figure 1 shown, the method for detecting the trajectory of a weak moving target based on a sonar image includes:

[0135] Step S11: Based on the fact that the gray level of the reverberation noise of the target detection sonar image follows a Weibull distribution, use the shape parameter, the scale parameter in the gray level, and a preset false alarm rate as the input parameters of the CA-CFAR detector.

[0136] In one implementation, based on the fact that the gray level of the reverberation noise of the target detection sonar image follows a Weibull distribution, using the shape parameter, the scale parameter in the gray level, and a preset false alarm rate as the input parameters of the CA-CFAR detector includes:

[0137] For the probability density function and the cumulative distribution function where x is a random variable, β is a shape parameter used to describe the aggregation degree of the reverberation data, η is a scale parameter used to describe the position point of the reverberation median, the probability density function of the mean statistic Y of the reference cell data is obtained as where Γ(θ) = ∫0 ∞ t θ-1 e -t dt, where θ is a function variable and Y is the mean statistic of the reference cell data;

[0138] Calculate the nominal factor through the false alarm rate equation in the CA-CFAR detector;

[0139] Through integral derivation, obtain from the nominal factor, where η is the scale parameter, PFA is the false alarm rate, and β is the shape parameter.

[0140] Step S12: Using a rectangular window as the selection criterion for the reference unit of the target detection sonar image, traverse each pixel value on the target detection sonar image to obtain the echo intensity accumulation value, and the rectangular window selects the reference unit in a sliding manner.

[0141] In one implementation, using a rectangular window as the selection criterion for the reference unit of the target detection sonar image, traverse each pixel value on the target detection sonar image to obtain the echo intensity accumulation value, and the rectangular window selects the reference unit in a sliding manner, including:

[0142] Using the rectangular window as the selection criterion for the reference unit of the target detection sonar image, starting from the center point of the rectangular window, select four trajectories at four angles of the cross shape at intervals of 45 degrees to accumulate the echo intensity values;

[0143] Starting from the center point of the rectangular window, form a range-azimuth two-dimensional reference window, accumulate the echo intensity values of a predetermined number of trajectories, sort them from small to large, and calculate the echo intensity accumulation value through the formula

[0144] where i and j are the abscissa and ordinate of the pixel point of the target to be detected in the target detection sonar image respectively, n is the number of motion trajectories, image(i,j) represents the echo intensity value of the pixel point (i,j), and track1 to trackn represent the 1st to nth motion trajectories respectively.

[0145] Step S13: When the echo intensity accumulation value is greater than or equal to the preset detection threshold, discard the trajectory with the minimum energy, splice the echo intensities on the remaining trajectories to obtain the corresponding reference unit; when the echo intensity accumulation value is less than the preset detection threshold, splice all the trajectories to obtain the corresponding reference unit.

[0146] Step S14: Estimate the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtain the detection threshold estimation value.

[0147] In one implementation, estimate the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtain the detection threshold estimation value, including:

[0148] Use the bisection method to solve the transcendental equation formed by the maximum likelihood function of the reference unit;

[0149] Iteratively calculate the shape parameter and scale parameter until it reaches a steady state to obtain the detection threshold estimation value.

[0150] Step S15: Compare the detection unit with the detection threshold estimation value to obtain a detection result.

[0151] Specifically, the detection unit is compared with the detection threshold estimation value, and if the condition is met, the detection result can be obtained.

[0152] The present invention proposes a HT-TBD (Hough transform-track before detection) detection method and system based on the trajectory of weak moving targets in sonar images, which includes the maximum accumulation of continuous frame sonar images, a CA-CFAR detector based on Weibull distribution, a reference window design of a two-dimensional detector of distance-azimuth, and parameter estimation of Weibull based on maximum likelihood estimation. The principle of the CA-CFAR detector based on Weibull distribution, the reference window design principle of the two-dimensional detector of distance-azimuth, and the principle and process of dual parameter estimation of Weibull distribution by the maximum likelihood estimation algorithm are described in detail.

[0153] I. About Weibull-CA-CFAR

[0154] According to a large number of studies, the clutter background in radar imaging obeys an exponential distribution, so the CFAR detection algorithm is mostly used in radar detection systems. For complex marine environments, especially the harsh seabed reverberation, weak targets in sonar images are a problem that needs to be solved urgently. The latest research shows that the grayscale pixel level of seabed reverberation obeys the Weibull distribution. Therefore, the present invention designs a target detector suitable for sonar images based on CA-CFAR. For the Weibull distribution, its probability density function and cumulative distribution function are expressed as follows:

[0155]

[0156] Among them, β is the shape parameter used to describe the aggregation of clutter data, and η is the scale parameter used to describe the location point of the clutter median. And the probability density function of the mean statistic Y of the reference unit data can be obtained as:

[0157]

[0158] in Referring to the principle of CA-CFAR detector again, and combining equations (4) and (8), we can obtain:

[0159]

[0160] Through further distribution integral derivation, the detection threshold T is finally obtained as:

[0161]

[0162] As can be seen from the above, if the shape parameter and scale parameter of the background noise distribution are known, the adaptive threshold can be calculated for each local area. However, in the underwater environment, the seabed reverberation is complex and variable, and the shape parameter and scale parameter are unknown. Therefore, it is necessary to estimate the noise parameters of the regional background noise to improve the robustness of the detection system. For this purpose, the present invention uses the maximum likelihood estimation (MLE) algorithm to estimate the parameters of the reference unit data in the sonar image, and the algorithm is deduced and the implementation process is carried out in detail in the next subsection.

[0163] II. MLE Estimation of Shape Parameter and Scale Parameter

[0164] From the probability density function of the Weibull distribution, and each reference unit is independent of each other, so the maximum likelihood function of N reference units can be obtained:

[0165]

[0166] Take the logarithm of both sides of the above formula, and let Finally, the expressions of the shape parameter and scale parameter are obtained as follows:

[0167]

[0168] It can be seen that the scale parameter is controlled by the shape parameter. However, equation (11) is a transcendental equation, and the bisection method is used to solve it. Set the solution range of the shape parameter and the absolute error ε, and let the initial interval length be l = β1 - β0. Then rewrite the transcendental equation of equation (12) into a unary nonlinear equation,

[0169]

[0170] Take the maximum value in the estimation range as the initial estimated value of β, iterate to calculate κ corresponding to the median of the range, and compare it with the remaining absolute error. If |κ n | - ε > 0, then use equation (15) to update the estimation range in the next iteration

[0171]

[0172] where n represents the number of rounds of iterative estimation, is the estimated value of the shape parameter. When |κ n | - ε ≤ 0, take the estimated as the final estimated value of the shape parameter. And substitute into equation (13) to obtain the estimated value of the scale parameter So far, the parameter estimation of Weibull is completed.

[0173] According to Part A and Part B, the CA-CFAR detector based on the Weibull distribution can be completely established. The weak target trajectory detection of sonar sequence images can be carried out by the Weibull-CA-CFAR detector.

[0174] III. Sonar Sequence Image Processing

[0175] Image Sequence Maximum Cumulation

[0176] The TBD moving target detection technology is not based on a single-frame image but on an image sequence. The image sequence forms a three-dimensional information space of range-azimuth-time, presenting a signal sequence within a certain range of range and azimuth for several time instants. And due to the limited moving speed of small underwater moving targets, their motion continuity characteristics in the three-dimensional space of range-azimuth-time are very obvious. Therefore, the sequence can be cumulated, and the trajectory detection can be carried out by using the continuity of target motion.

[0177] The sonar image sequence can be expressed as f(r,θ,t i ), i = 1, 2,..., n, where n represents the number of frames of the sonar image. There are three cumulative methods: single-frame detection followed by cumulation, summation cumulation followed by detection, and maximum cumulation followed by detection. However, according to the paper "Research on Detection and Tracking Technology of Small Underwater Moving Targets Based on Co-located MIMO Image Sonar", the form of single-frame detection followed by cumulation loses the target motion characteristics, and the method of summation cumulation followed by detection greatly damages the image signal-to-noise ratio. While the method of maximum cumulation followed by detection can better suppress clutter with little loss of the image signal-to-noise ratio and shows better performance. Therefore, the present invention selects the form of maximum cumulation followed by detection to process the sonar sequence images, and its expression form is as follows:

[0178] g(r,θ) = max(f(r,θ,t1), f(r,θ,t2),..., f(r,θ,t n )) (16)

[0179] The maximum cumulation of sonar sequence images needs to traverse all range-azimuth information within the time range. Through maximum cumulation, the three-dimensional information of range-azimuth-time of the sonar image is mapped onto the range-azimuth plane, and the constructed detector is used to detect the cumulative image to obtain the trajectory detection result of the sequence image.

[0180] Range-Azimuth Two-Dimensional CFAR Reference Window

[0181] Prior to this, the traditional CA-CFAR selected a rectangular window in the detection of images. However, using a rectangular window to obtain reference cells cannot well obtain the background features in azimuth and range. For this reason, some researchers proposed using a cross window to select reference cells, which makes better use of the range and azimuth information. However, since the target trajectory has a scale when moving in a certain direction, such as the target moving horizontally or vertically, using a cross window to select reference cells will increase the background estimation value and increase the detection threshold, resulting in missed detection of small targets with low signal-to-noise ratio. For this reason, according to the motion characteristics of the target and the fact that the seabed reverberation has no motion characteristics, the present invention designs a reference cell selection method applicable to the detection of underwater moving target trajectories, and sets different false alarm rates for each different type of detection cell, which more effectively suppresses clutter and improves the trajectory detection effect of weak small targets.

[0182] In the selection of reference cells for two-dimensional CFAR detection, the selection of the target trajectory should be avoided. Since the underwater target has a limited movement speed and moves in a straight line in a short time, the present invention selects 4 prediction trajectories within the rectangular window. As Figure 7 shown,

[0183] The echo intensity values of the four trajectories are accumulated and sorted from small to large as

[0184]

[0185] where i and j respectively represent the range and azimuth coordinate values of the sonar cumulative image. After sorting the cumulative values of the prediction trajectories, in order to determine whether the trajectory with the largest cumulative value contains the target trajectory, an energy ratio threshold σ is set. The following formula is used to determine whether the predicted trajectory contains the target trajectory:

[0186]

[0187] If the predicted trajectory satisfies inequality (18), it is considered that the prediction trajectory with the largest cumulative sum contains the target motion trajectory. Therefore, the reference cells on the other three trajectories are selected for splicing as reference data. If the above conditions are not met, it is considered that the area near the target is a uniform clutter environment, and the reference cells on the above four trajectories are selected for splicing as reference data. After traversing all range-azimuth cells, each cell obtains corresponding reference data. For the detection cell that may contain the target trajectory, a false alarm rate P f1 is set, and a false alarm rate of P f2 is set for the detection cell with a uniform clutter environment nearby. Among them, according to the preliminary judgment of the existence of the trajectory, to improve the suppression effect of false speckle noise, the set false alarm rate satisfies P f2 <<P f1 .

[0188] IV. Sonar Sequential Image Detection System

[0189] Based on the above reasoning and analysis, the present invention proposes a trajectory detection system model for sonar sequential images based on HT-TBD of Weibull distribution.

[0190] As Figure 8 shown, in the moving target trajectory detection system of sonar sequential images, the sonar sequential images are processed through a maximum accumulation strategy. The N-frame sonar images {image_1, image_2,..., image_N} are used to calculate the multi-frame joint image of distance-azimuth through Equation (16). Then, four prediction trajectories are selected within the reference window of each cell to be detected in the joint image. The corresponding reference cell data are selected through Equation (18), and the shape parameter and scale parameter of the background noise are estimated by the bisection method using the principle of the maximum likelihood estimation (MLE) algorithm. Furthermore, the false alarm rate is set according to the different reference cells selected for each cell to be detected. The adaptive threshold of each cell to be detected is obtained using the detection principle of CA-CFAR, and the final detection result of the multi-frame joint image is obtained through a comparator. Since there are many false strong speckle noises in the seabed reverberation, the detection result is further processed by median filtering, and finally, the Hough transform line detector is used to perform line detection on the processing result of the above multi-frame joint image to obtain the trajectory of the moving target.

[0191] Figure 2 Fig. shows a flowchart of another method for detecting the trajectory of a weak moving target based on sonar images according to an embodiment of the present invention. As Figure 2 shown, the method for detecting the trajectory of a weak moving target based on sonar images can be generally divided into:

[0192] Step S21: Obtain multiple frames of sonar images of the moving target, and perform a joint operation on the multiple frames of sonar images through the maximum accumulation algorithm to obtain a detection image of the moving target;

[0193] Step S22: Detect the target point traces in the detection image using the aforementioned method;

[0194] Step S23: Determine the moving trajectory of the moving target through a trajectory detector.

[0195] In one implementation, obtaining multiple frames of sonar images of the moving target and performing a joint operation on the multiple frames of sonar images through the maximum accumulation algorithm to obtain a detection image of the moving target includes:

[0196] Traverse multiple frames of sonar images of the moving target within a specified time range, analyze the distance-azimuth information, and construct a three-dimensional information model of distance-azimuth-time;

[0197] Perform the maximum accumulation algorithm on the three-dimensional distance-azimuth-time information model to obtain a detection image of the moving target.

[0198] In one implementation, determining the motion trajectory of the moving target through a trajectory detector includes:

[0199] Perform median filtering on the detection result to obtain a median filtering result;

[0200] Perform Hough transform on the median filtering result to obtain a trajectory detection result.

[0201] Figure 3 Show the composition diagram of a weak moving target trajectory detection system based on sonar images according to an embodiment of the present invention. As Figure 3 shown, the weak moving target trajectory detection system based on sonar images as a whole can be divided into:

[0202] An input module 31, configured to, based on the fact that the gray level of the reverberation noise of the target detection sonar image follows a Weibull distribution, use the shape parameter, the scale parameter in the gray level, and a preset false alarm rate as input parameters of the CA-CFAR detector;

[0203] A selection module 32, configured to use a rectangular window as the selection reference for the reference unit of the target detection sonar image, traverse each pixel value on the target detection sonar image to obtain an echo intensity accumulation value, and the rectangular window selects the reference unit in a sliding form;

[0204] A judgment module 33, configured to, when the echo intensity accumulation value is greater than or equal to a preset detection threshold, discard the trajectory with the minimum energy, splice the echo intensities on the remaining trajectories to obtain a corresponding reference unit; when the echo intensity accumulation value is less than the preset detection threshold, splice all the trajectories to obtain a corresponding reference unit;

[0205] An estimation module 34, configured to estimate the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtain a detection threshold estimation value;

[0206] A comparison module 35, configured to compare the detection unit with the detection threshold estimation value to obtain a detection result.

[0207] Figure 4 Show the composition diagram of another weak moving target trajectory detection system based on sonar images according to an embodiment of the present invention. As Figure 4 shown, the weak moving target trajectory detection system based on sonar images as a whole can be divided into:

[0208] The maximum accumulation module 41 is configured to obtain multiple frames of sonar images of a moving target, perform a joint operation on the multiple frames of sonar images through a maximum accumulation algorithm, and obtain a detection image of the moving target;

[0209] The detection module 42 is configured to perform target point detection on the detection image by using the foregoing method;

[0210] The result module 43 is configured to determine the motion trajectory of the moving target through a trajectory detector.

[0211] Figure 5 The composition diagram of the estimation module according to an embodiment of the present invention is shown. As Figure 5 shown, the estimation module 34 may include:

[0212] The solution unit 341 is configured to solve the super equation formed by the maximum likelihood function of the reference unit by using the bisection method;

[0213] The iteration unit 342 is configured to iteratively calculate the shape parameter and the scale parameter until reaching a steady state, and obtain an estimated value of the detection threshold.

[0214] Experimental verification

[0215] During the verification process, the data set used in the present invention is URPC2021, which is launched by Pengcheng Laboratory, and the website address is https: / / code.ihub.org.cn / projects / 14186. These data are obtained by a sonar of the Tritech model, which has detected eight static targets in the real ocean. And the data are collected together with the water depth detection and seabed detection data. Since the present invention is based on the trajectory detection of small underwater moving targets, only the seabed reverberation background part of the actual data is intercepted for experiments. And this experiment is tested on a CPU of the model Intel(R) Core(TM) i7-10710U.

[0216] The shared parameters set in this verification experiment are: the number of frames N of the sonar image sequence is selected as 10 frames, the number of predicted trajectories n is 4, the threshold of the mean ratio of track energy σ is 3, the maximum likelihood estimation accuracy ε is 0.0001, and the false alarm rate P f1 is 0.1, and the false alarm rate P f2 is 0.001.

[0217] A Sonar image sequence image data construction

[0218] In this experiment, since the range of the measured ocean data is not wide, it is assumed that the change in the target echo energy intensity between each frame is not significant, and there is only a change in position. Therefore, by setting different displacement amounts of the target in the range-azimuth, the true velocity value of the target can be obtained by using Equation (19):

[0219]

[0220] Where v is the actual speed of the obtained target; s is the actual maximum distance detected by the forward-looking sonar; d is the pixel number displacement of the target between frame images; fps is the frame refresh rate of the forward-looking sonar; H is the number of pixels in the distance dimension corresponding to the forward-looking sonar image; v′ is the unit relationship between the speed and the underwater speed. By setting different pixel displacements d, the actual speed of the target can be obtained and compared with the current actual underwater target movement speed to ensure the reliability of the constructed sonar sequence images. For example, the Ganendra RI-1 series DPV (diver propulsion vehicle) can achieve a thrust of 50 kg, and the maximum speed can reach 3.5 kN; and the speed of the bluefin series UUV is in the range of 0.5 - 5 kN.

[0221] Figure 9 visualizes the result of the maximum accumulation of ten frames of forward-looking sonar images. The vertical axis represents the distance dimension, and the horizontal axis represents the direction dimension. (a), (b), and (c) indicate that the detected targets move in different directions, and the initial positions of the targets are random. The target movement displacement between frames is set to 8 pixels. According to the measured background depth of 13.0012 m, the height of the sonar image is 1595 pixel points, and the frame refresh rate is 30 fps. Therefore, the constant speed of the target movement can be obtained as 3.8 kN through Equation (19).

[0222] B. Verification of Measured Seafloor Reverberation Distribution

[0223] To further verify that the seafloor reverberation follows the Weibull distribution, the following is a comparative analysis and verification through parameter estimation and simulation of the measured seafloor reverberation.

[0224] As shown in Figure 10, Figure 10a shows the measured reverberation sonar image. The Gaussian distribution, Rayleigh distribution, gamma distribution, and Weibull distribution are used to fit the gray histogram of the seafloor reverberation. The fitting effects of various distributions are as Figure 10b shown.

[0225] We use the Kolmogorov distance criterion and the χ 2 criterion to quantitatively evaluate the fitting results. The results are shown in Table 1. The evaluation function of the Kolmogorov distance criterion is defined as follows:

[0226]

[0227] Where h is the normalized sonar image histogram, p is the fitting curve, and L is the gray level of the image. The standard definition of the χ 2 criterion is as follows:

[0228]

[0229]

[0230] According to the above data analysis, the Weibull distribution has the minimum error in both the Kolmogorov distance and the χ 2 distance. Therefore, the gray pixel level of the seabed reverberation is closest to the Weibull distribution. This result verifies the effectiveness of our algorithm.

[0231] C. Underwater moving target detection effect

[0232] Figure 11 shows the seventh frame image captured when the target is moving horizontally. As Figure 11c shown, the target echo is weak and submerged in the complex seabed reverberation. It is impossible to directly extract the position of the target from a single frame. Based on the principle of traditional detection and tracking (DBT), if the set detection threshold is less than the target echo intensity at this time, the target can be detected, but there are still many false strong echo noises. If the set detection threshold is greater than the target echo intensity, it will lead to detection missed alarms. Therefore, in such a complex seabed reverberation environment, it is impossible to use the traditional detection and tracking method after detection to detect and track the target.

[0233] In addition, due to the weak target intensity and the uneven measured environment, etc., the local signal-to-noise ratio (LSNR) of the image is used to define the SNR (signal-to-noise ratio) of small targets in the sonar image. Its definition expression is as shown in formula (20):

[0234]

[0235] Among them, E r is the mean value of the target area, E B is the mean value of the background area, and δ B is the standard deviation of the background area. Therefore, the following several TBD algorithms are used to detect and test this forward-looking sonar sequence image. And the detection performance of the target moving in three directions: horizontal, vertical, and diagonal is analyzed.

[0236] As shown in Figure 12, the present invention uses a variety of algorithms to detect the sonar image sequence in which the target moves horizontally. Figure 12aShows the results of size selection and accumulation of an image sequence of the target's horizontal movement. The white box represents the moving trace of the target. Statistically, the pixel gray level of the strongest echo in the seabed reverberation collected in ten frames reaches 140. Here, the pixel gray level of the echo intensity of the target is set to 40. And the regional signal-to-noise ratio of the target in each frame is different, which is affected by the non-uniformity of the reverberation. In the detection experiment of the horizontal movement trajectory, it is assumed that the initial position of the target is (150, 160), where 150 represents the pixel coordinate in the range dimension and 160 represents the coordinate in the azimuth dimension, and the moving speed is 3.8 kn.

[0237] Figure 12b Shows the detection results of the traditional CA-CFAR detector. A rectangular window is used to select the reference unit, where the length and height of the rectangular window are 13, and the length and height of the guard window are 5. Figure 12c The reference window type used for CA-CFAR detection in [reference] is the cross type, where the length of the reference window is 8 and the length of the guard window is 4. Figure 12d Shows the results of detecting the forward-looking sonar sequence images by manually setting the threshold, where the threshold is set to the critical point where the target can be detected. Figure 12e Is the ideal result of the image sequence, only a schematic diagram of the target echo. Figure 12f Shows the detection results of the Weibull-CA-CFAR detector with a rectangular reference window, where the size of the reference window is consistent with the result in Fig. 12(b). Figure 12g Shows based on Figure 12a The CA-CFAR detection results of the estimation of the two parameters of Weibull in [reference], which is the estimation result of a global threshold. Figure 12h Shows the detection results of the detection algorithm of the present invention.

[0238] The experimental analysis is as follows. Figure 12b Using the traditional CA-CFAR detection method is not suitable for detecting moving targets in the seabed reverberation environment. The detection results not only have most of the targets missed alarms, but also the seabed reverberation suppression effect is poor. In Figure 12c A cross window is used to select the reference unit. Although it can better extract the background noise information in the azimuth dimension and improve the reverberation suppression effect to a certain extent, there are still a large number of missed alarms. Figure 12d Using the method of manually setting the threshold. This method is more practical in actual methods. DBT is also based on this method, but it is practical in the case of strong target echo and high signal-to-noise ratio. For sonar images with complex backgrounds, due to the existence of a large number of false speckle noises and weak target echoes, it is difficult to accurately set the detection threshold. In addition, while ensuring that the target is not missed in the alarm, there will be a lot of strong echo noises in the detection results, affecting further judgment. Figure 12eShows the ideal detection result. All reverberations are suppressed and the detection target is left. In Figure 12f , a fixed false alarm rate is set. Although the pixel distribution of seabed reverberation is introduced, a higher false alarm rate will be generated for an environment with a large amount of speckle noise and unevenness. In addition, the selection of the reference unit does not consider the movement trajectory of the target, so the detection effect is still not good, but it is improved compared with the traditional CA-CFAR. Figure 12g Shows the result of global threshold estimation. The principle is to use Weibull-CA-CFAR to estimate the Figure 12a detection threshold in. Compared with Figure 12d , this method does not require manual threshold setting and realizes a certain degree of automation. However, due to the limitation of the global threshold, there are serious false speckle noises in the detection result. Figure 12h Shows the detection algorithm proposed by the present invention. As can be seen from the above figure, the seabed reverberation parameter estimation based on the Weibull distribution and the improved reference unit design strategy proposed by the present invention have better detection effects than the earlier algorithms and significantly improve the reverberation suppression effect. Through analysis, the detection effect of this algorithm is the best.

[0239] Shown in Figure 13 is the result of detecting the straight-line trajectory of the target using the Hough transform for all the detection results in Figure 12. As shown in the figure, when the highest gray value of the overall background is 140, the gray pixel of the target is 40, and the signal-to-noise ratio of the target area is weak, the trajectory detection effect of the present invention reaches the best, and the movement trajectory of the target obtained by collecting these ten frames can be detected. Compared with the other five algorithms, due to the influence of reverberation interference, there are a large number of missed alarm phenomena or detection error phenomena.

[0240] As shown in Figure 14, the detection of sonar sequence images in which the target moves vertically by various algorithms is visible. Figure 14a Among them, the white box represents the moving trace of the target. Among them, the relevant parameters of the target are set for the sequence image of the forward-looking sonar. The strongest echo pixel value of the reverberation is 140, the set target strength is 50, the initial position is (250, 200), and the movement speed is 3.8 kN.

[0241] From the detection results of each algorithm, it can be seen that the algorithm of the present invention has a better reverberation suppression effect than the other five algorithms, and has a higher detection rate, and can clearly distinguish the movement trajectory of the target. As can be seen from the trajectory detection result of the detection result in Figure 15, Figure 15f , 15gThe algorithm that estimates the regional seabed reverberation noise by referring to the Weibull distribution for 15h and then calculates the adaptive threshold can effectively detect the target's motion trajectory, and has better performance compared with several other traditional CA-CFAR detection algorithms, and can be well applied to forward-looking sonar images. However, compared with the algorithm of the present invention, the strategies of selecting reference cells using a rectangular window and calculating the global threshold show poor performance, and the algorithm of the present invention shows more excellent performance in reverberation suppression.

[0242] Figure 16 shows the detection results of various algorithms for the target moving in a diagonal direction. Figure 16a In it, the white box represents the moving trace of the target. Set the initial position of the target to (250, 160), the moving speed to 5.4 kN, and the pixel value of the target echo intensity to 40. By analyzing the detection results, it can be seen that due to the random setting of the position, when the reverberation intensity around the target is strong and changes violently, it will lead to a weak regional signal-to-noise ratio, resulting in a poor detection effect. As can be seen from Figure d, when setting the detection critical threshold of the target, while detecting the target, there are also a large number of strong reverberation noises around the target. So in Figure 16g the algorithm of the present invention suppresses the target to a certain extent while suppressing the reverberation, so the detection effect is slightly worse than the detection effects of the above horizontal movement and vertical movement. However, in the horizontal comparison, the algorithm of the present invention still has the best detection effect. As shown in Figure 17, through the Hough transform, the motion trajectory of the target can be displayed, and at the same time, the suppression effect on the reverberation noise is the best.

[0243] By setting the target motion models in three directions, the initial position of each case is random, and the target echo intensity is weak, and then each algorithm is tested. Observing and analyzing the detection results, the following can be summarized: In different cases, the algorithm of the present invention can detect the target's motion trajectory, and has the best suppression effect on the seabed reverberation, further verifying the generalization of the algorithm of the present invention.

[0244] In order to better evaluate the performance among various algorithms, the detection rates under different signal-to-noise ratios are compared and analyzed. Since the environment is non-uniform and the regional signal-to-noise ratio cannot be used as a variable, the target echo intensity is used instead of the regional signal-to-noise ratio as a variable in the experiment for the performance test of the detection rate. The motion trajectories in the horizontal, vertical, and diagonal directions at eight speeds of 1.9 kN, 2.4 kN, 2.9 kN, 3.4 kN, 3.8 kN, 4.3 kN, 4.8 kN, and 5.3 kN are tested, where each speed is tested ten times, and the position of each movement is random. Through each algorithm for trajectory detection, the mean value of the 240 detection rates under each echo intensity is calculated as the final detection rate under that target echo intensity.

[0245] AsFigure 18 As shown, by comparing the detection performances of different algorithms, the algorithm proposed by the present invention has the best detection performance. When the detection rate of this method reaches 100%, at the same target intensity, it is 28.28%, 11.67% and 5.87% higher than the cross-window method CA-CFAR, the rectangular window method CA-CFAR and the rectangular window method Weibull-CA-CFAR respectively. In addition, when the target gray level is 53, the detection rate of this algorithm can reach 75%. However, the CA-CFAR using the cross-window method requires 80 gray level values, the CA-CFAR using the rectangular window method requires 65 gray level values, and the Weibull-CA-CFAR using the rectangular window method requires 60 gray level values.

[0246] The present invention proposes a moving target trajectory detection system based on HT-TBD for the moving target detection task of forward-looking sonar images.

[0247] First, the forward-looking sonar image sequence is processed by adopting the strategy of maximum accumulation. Secondly, based on the CA-CFAR detection algorithm, the distribution of seabed reverberation is introduced into the detection principle, and thus a moving target detector for forward-looking sonar images based on Weibull-CA-CFAR is proposed. Furthermore, aiming at the phenomena of serious speckle noise and uneven noise in the reverberation, and in order to avoid the influence of the target trajectory, an improved reference cell selection strategy is proposed, which can eliminate the influence of target traces in the reference cells. Then, the improved reference cell selection strategy is introduced into the improved Weibull-CA-CFAR algorithm, and the maximum accumulation result is detected to obtain the trace detection result. Finally, further processing is carried out through median filtering, and then trajectory detection is carried out on the trace detection result through Hough transform.

[0248] Through the comparison of algorithm performances, the improved Weibull-CA-CFAR detection algorithm proposed by the present invention has the best detection accuracy and seabed reverberation suppression effect. However, the TBD detection algorithm based on the CFAR detector has a large amount of calculation. This method needs to estimate the threshold for each cell, and the parameter estimation of the noise has a certain degree of complexity. Therefore, the next step should focus on reducing the complexity of the algorithm to better complete the underwater moving target early warning task.

[0249] For the functions of each module in each system of the embodiments of the present invention, reference can be made to the corresponding descriptions in the above method, which will not be elaborated here.

[0250] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0251] Any process or method description shown in the flowchart or described in other ways herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions may be executed in a manner not shown or discussed, including in a substantially simultaneous manner according to the functions involved or in a reverse order, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0252] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in connection with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection portion having one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0253] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques known in the art or a combination thereof can be used: discrete logic circuits having logic gate circuits for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0254] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the method of the above embodiments can be completed by instructing relevant hardware through a program. The said program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0255] In addition, each functional unit in various embodiments of the present invention can be integrated in a processing module, or each unit can exist physically alone, or two or more units can be integrated in a module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk, an optical disc, etc.

[0256] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various changes or substitutions thereof, and these should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the said claims.

Claims

1. A method for detecting the trajectory of a weak moving target based on sonar images, characterized in that Including: Based on the fact that the gray level of reverberation noise in the target detection sonar image follows a Weibull distribution, the shape parameter, scale parameter in the gray level, and a preset false alarm rate are used as the input parameters of the CA-CFAR detector; Taking a rectangular window as the selection criterion for the reference unit of the target detection sonar image, traversing each pixel value on the target detection sonar image to obtain the cumulative value of the echo intensity, and the rectangular window selects the reference unit in a sliding form; When the cumulative value of the echo intensity is greater than or equal to the preset detection threshold, discard the trajectory with the minimum energy, and splice the echo intensities on the remaining trajectories to obtain the corresponding reference unit; When the cumulative value of the echo intensity is less than the preset detection threshold, splice all the trajectories to obtain the corresponding reference unit; Estimate the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtain the detection threshold estimation value; Compare the detection unit with the detection threshold estimation value to obtain the detection result.

2. The method according to claim 1, characterized in that Based on the fact that the gray level of reverberation noise in the target detection sonar image follows a Weibull distribution, using the shape parameter, scale parameter in the gray level, and a preset false alarm rate as the input parameters of the CA-CFAR detector, including: For the probability density function of the Weibull distribution and the cumulative distribution function where x is a random variable, β is a shape parameter used to describe the clustering degree of reverberation data, and η is a position point scale parameter used to describe the median of reverberation, the probability density function of the mean statistic Y of the reference cell data is where where θ is a function variable and Y is the mean statistic of the reference cell data; Calculate the nominal factor through the false alarm rate equation in the CA-CFAR detector; Obtained from the nominal factor through integral derivation where η is the scale parameter, PFA is the false alarm rate, and β is the shape parameter.

3. The method according to claim 1, characterized in that Taking a rectangular window as the selection criterion for the reference unit of the target detection sonar image, traversing each pixel value on the target detection sonar image to obtain the cumulative value of the echo intensity, and the rectangular window selects the reference unit in a sliding form, including: Taking the rectangular window as the selection criterion for the reference unit of the target detection sonar image, starting from the center point of the rectangular window, select four trajectories at four angles in a cross shape at intervals of 45 degrees to accumulate the echo intensity values; Starting from the center point of the rectangular window, a range-azimuth two-dimensional reference window is formed, the echo intensity values of a predetermined number of trajectories are accumulated, sorted from small to large, and through the formula Calculate the cumulative value of the echo intensity, where i and j are the abscissa and ordinate of the pixel point of the target to be detected in the target detection sonar image respectively, n is the number of motion trajectories, image(i,j) represents the echo intensity value of the pixel point (i, j), and track1 to trackn respectively represent the 1st to nth motion trajectories.

4. The method according to claim 1, wherein Estimate the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtain the detection threshold estimation value, including: Use the bisection method to solve the transcendental equation formed by the maximum likelihood function of the reference unit; Iteratively calculate the shape parameter and scale parameter until it reaches a steady state to obtain the detection threshold estimation value.

5. A method for detecting the trajectory of a weak moving target based on sonar images, characterized in that, Including: Obtain multiple frames of sonar images of a moving target, perform a joint operation on the multiple frames of sonar images through the maximum accumulation algorithm to obtain a detection image of the moving target; Use the method described in claim 1 to perform target point detection on the detection image; Determine the motion trajectory of the moving target through a trajectory detector.

6. The method according to claim 5, characterized in that, Obtain multiple frames of sonar images of a moving target, perform a joint operation on the multiple frames of sonar images through the maximum accumulation algorithm to obtain a detection image of the moving target, including: Traverse multiple frames of sonar images of the moving target within a specified time range, analyze the range-azimuth information, and construct a three-dimensional information model of range-azimuth-time; Perform a maximum cumulative algorithm on the three-dimensional information model of range-azimuth-time to obtain a detection image of the moving target.

7. The method according to claim 5, wherein Determine the motion trajectory of the moving target through a trajectory detector, including: Perform median filtering on the detection result to obtain a median filtering result; Perform Hough transform on the median filtering result to obtain a trajectory detection result.

8. A weak moving target trajectory detection system based on sonar images, characterized in that, Including: An input module, which, based on the fact that the gray level of the reverberation noise of the target detection sonar image follows a Weibull distribution, uses the shape parameter, scale parameter in the gray level, and a preset false alarm rate as the input parameters of the CA-CFAR detector; A selection module, which uses a rectangular window as the selection criterion for the reference unit of the target detection sonar image, traverses each pixel value on the target detection sonar image, and obtains the cumulative value of the echo intensity. The rectangular window selects the reference unit in a sliding manner; A judgment module, which, when the cumulative value of the echo intensity is greater than or equal to a preset detection threshold, discards the trajectory with the minimum energy, splices the echo intensities on the remaining trajectories, and obtains the corresponding reference unit; When the cumulative value of the echo intensity is less than the preset detection threshold, splice all the trajectories to obtain the corresponding reference unit; An estimation module, which estimates the reference unit through maximum likelihood estimation to obtain the shape parameter and scale parameter of the reverberation noise distribution, and obtains an estimated value of the detection threshold; A comparison module, which compares the detection unit with the estimated value of the detection threshold to obtain a detection result.

9. The system according to claim 8, wherein The estimation module includes: A solution unit, which uses the bisection method to solve the transcendental equation formed by the maximum likelihood function of the reference unit; An iteration unit, which iteratively calculates the shape parameter and scale parameter until it reaches a steady state to obtain an estimated value of the detection threshold.

10. A faint moving target trajectory detection system based on sonar images, characterized in that, Including: A maximum cumulative module, which obtains multiple frames of sonar images of the moving target, and performs a joint operation on the multiple frames of sonar images through the maximum cumulative algorithm to obtain a detection image of the moving target; A detection module, which uses the method described in claim 1 to perform target point detection on the detection image; A result module, which determines the motion trajectory of the moving target through a trajectory detector.

Images