Sonar image target rapid constant false alarm rate detection method based on segmentation ordered weighting

Abstract

The invention discloses a rapid constant false alarm detection method for a sonar image target based on segmentation ordered weighting, and relates to the technical field of sonar image target detection. The detection method comprises the following steps: 1, segmenting a target potential area; step 2, carrying out local constant false alarm rate detection on the segmented area; the framework of the invention adopts a two-stage method: firstly, a target area is segmented by using an Otsu method global threshold value, and then local advanced two-dimensional constant false alarm rate detection is carried out; in a high-resolution sonar image in which a target occupies few pixels, the Otsu method segmentation effectively eliminates a non-target area and an obvious interference source which may trigger a false alarm, so that the calculation complexity and the false alarm rate of a subsequent detection process are remarkably reduced; when two-dimensional constant false alarm rate detection is carried out locally, two kinds of ordered weighting coefficients are fused and used for a reference unit, a rectangular weighting coefficient is adopted to suppress strong interference under the condition of a high signal-to-noise ratio, and a trapezoidal weighting coefficient is adopted to prevent missing detection under the condition of a low signal-to-noise ratio.

Description

Technical Field

[0001] The present invention belongs to the technical field of sonar image target detection, and particularly relates to a method for fast constant false alarm detection of sonar image targets based on segmented ordered weighting. Background Art

[0002] Imaging sonar is one of the most important high-tech marine detection devices in the contemporary marine engineering field, and is widely used in fields such as subsea engineering construction and maintenance, marine mapping, and marine resource exploration. In the actual detection of imaging sonar, affected by the complex and variable marine sound field, environmental noise, and unstable motion state of the carrier, the target echo signal-to-noise ratio is relatively low, there are many speckle noises in the sonar image, and the imaging quality is very unstable. Reducing the false alarm rate and the probability of missed detection in such a complex environment and reliably detecting the target is an important issue in the field of sonar signal processing. Constant False Alarm Rate (CFAR) detection technology has been proven to have good ability to control false alarms in a background that follows a Gaussian distribution. With the continuous development of CFAR detection technology, this technology has been widely applied in the field of sonar technology.

[0003] How to maintain the best detection performance in the complex detection background of sonar images is an important issue. The computational complexity of advanced detection algorithms brings huge processing overhead. Comprehensive detection and processing of the entire frame of sonar image requires a large amount of computing resources and processing time, so it is necessary to develop effective detection strategies to enhance the real-time processing ability in sonar image target detection. Summary of the Invention

[0004] In order to solve the existing problems, the purpose of the present invention is to provide a method for fast constant false alarm detection of sonar image targets based on segmented ordered weighting.

[0005] The method for fast constant false alarm detection of sonar image targets based on segmented ordered weighting of the present invention is as follows:

[0006] Step 1: Adopt a segmentation-detection framework, combine global Otsu segmentation and local two-dimensional CFAR detection. Before local detection, first use the global Otsu segmentation algorithm to reduce the computational complexity and false alarm rate in the subsequent detection process;

[0007] Step 2: Adopt the Segmented-Ordered-Weighting CFAR (SOW-CFAR) detection method. After segmenting the potential target area, in local two-dimensional CFAR detection, use rectangular and trapezoidal ordered weighting coefficients for reference unit fusion, improve the adaptability of the detection algorithm to the complex detection background of sonar images, reduce the probability of missed detection, and enhance the detection performance.

[0008] Preferably, in the segmentation-detection framework in the first step, a global segmentation threshold is first calculated according to the different distribution characteristics of the two types of data, namely the target and the background, to quickly segment the target area; then, a local adaptive threshold is calculated using a two-dimensional CFAR sliding detection window in the segmented target area to achieve precise detection of the target.

[0009] Preferably, for the segmentation ordered weighted constant false alarm detection method, the area where the target is located is first segmented using the Otsu method, and then two-dimensional ordered weighted constant false alarm detection is performed based on a two-dimensional detection window that takes into account the noise in both the azimuth and range directions.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0011] First, the framework adopts a two-stage method: first, the Otsu method is used for global threshold segmentation to segment the target area, and then local advanced two-dimensional CFAR detection is performed; in high-resolution sonar images where the target occupies very few pixels, the Otsu segmentation effectively eliminates non-target areas and obvious interference sources that may trigger false alarms, thus significantly reducing the computational complexity and false alarm rate in the subsequent detection process.

[0012] Second, on the basis of the segmentation-detection framework, a SOW-CFAR detection method is further proposed; in the local detection stage, a hybrid method combining rectangular and trapezoidal ordered weighting coefficients is adopted. In the case of high signal-to-noise ratio, rectangular weighting coefficients are used to suppress strong interference, and in the case of low signal-to-noise ratio, trapezoidal weighting coefficients are used to prevent missed detections. This weighting strategy enhances the robustness of the algorithm in complex sonar imaging backgrounds and improves the overall detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] For ease of explanation, the present invention is described in detail by the following specific embodiments and accompanying drawings.

[0014] Figure 1 Schematic diagram of the structure of a one-dimensional CFAR detector.

[0015] Figure 2 Schematic diagram of the segmentation-detection framework;

[0016] Figure 3 Schematic diagram of the principle of local two-dimensional ordered weighted CFAR detection when a cross detection window is used in SOW-CFAR;

[0017] Figure 4 Schematic diagram of a water body image collected by a multibeam echosounder;

[0018] Figure 5 Schematic diagram of another water body image collected by a multibeam echosounder;

[0019] Figures 6 to 9 For Figure 4Schematic diagram of the detection results of the multi-beam sounding water body image; among which Figure 6 is CA, Figure 7 is SCA, Figure 8 is OS, Figure 9 is SOS;

[0020] Figures 10 to 12 is for Figure 5 Schematic diagram of the detection results of the multi-beam sounding water body image; among which Figure 10 is CA, Figure 11 is OS, Figure 12 is SOW. Specific implementation manners

[0021] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be described below through specific embodiments shown in the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. The structures, ratios, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those skilled in this technology to understand and read, and are not used to limit the implementation conditions of the present invention. Therefore, they do not have technical substance. Any modification of the structure, change of the ratio relationship or adjustment of the size, without affecting the effects that the present invention can produce and the objectives that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0022] Here, it should also be noted that in order to avoid obscuring the present invention due to unnecessary details, only the structures and / or processing steps closely related to the solution according to the present invention are shown in the drawings, and other details less related to the present invention are omitted.

[0023] I. CFAR detection:

[0024] As Figure 1 shown, the segmentation-detection framework is derived from the one-dimensional CFAR theory. To better analyze the segmentation-detection framework, the structural schematic of the one-dimensional CFAR detector is as Figure 1 shown.

[0025] The CFAR detector is based on a sliding window structure, forming an adaptive threshold locally, and can detect data with a relatively constant false alarm rate under ideal conditions. In this embodiment, D is used to represent the detection statistic in the detection unit; adjacent to the detection unit are two protection units, which are used to prevent the target energy from leaking into the reference unit and affecting the local estimation of the clutter intensity; x i(i = 1, ..., 2n) represents the reference statistics in the reference cells on both sides of the detection unit; the reference sliding window length is set to R = 2n, where n is the unilateral reference sliding window length; Z represents the local estimate of the clutter intensity in the sliding reference window formed by x i The clutter intensity local estimate in the sliding reference window formed by x

[0026]

[0027] In the formula, H1 represents the hypothesis of the presence of a target, and H0 represents the hypothesis of the absence of a target.

[0028] The envelope of Gaussian distributed clutter follows a Rayleigh distribution. After square-law detection, each reference cell sample in the detection window follows an exponential distribution, and its probability density function is:

[0029]

[0030] Under the hypothesis H0 of no target in the reference cells, λ' is the total average power level of the background clutter plus thermal noise, denoted by μ; under the hypothesis H1 of the presence of a target, λ' is μ(1 + SNR). Where SNR is the average power ratio of the target signal to the clutter plus noise, so

[0031]

[0032] Assume that the reference cell samples are independent and identically distributed, and their λ' are all μ; since the threshold S = TZ is a random variable, it is necessary to take the statistical average of Z, and the false alarm probability is expressed as:

[0033]

[0034] Where, f Z (z) is the probability density function of Z, and M Z (u) is the moment generating function (MGF) of Z. When This formula is transformed into the detection probability expression in a uniform clutter background:

[0035]

[0036] For the cell-averaging CFAR (CA-CFAR) detection method, the estimate of the background power level is obtained from the mean of R reference cell samples, that is:

[0037]

[0038] Where x iIt obeys the exponential distribution. Obviously, the exponential distribution is a special gamma distribution when n=1, that is, x i ~G(1,1 / μ). Since the reference unit samples are independent and identically distributed, Z CA ~G(R,R / μ), so Z CA The moment generating function of

[0039]

[0040] Using the moment generating function, the false alarm probability and detection probability of CA-CFAR can be obtained as follows:

[0041]

[0042] For ordered-statistics CFAR (OS-CFAR), the R reference unit samples must first be sorted from small to large:

[0043] x (1) ≤x (2) ≤…≤x (R) (10)

[0044] Then take the kth sorted sample as the estimate Z of the background power level, that is:

[0045] Z OS =x (k) (11)

[0046] Z OS The probability density function can be expressed as:

[0047]

[0048] So Z OS The moment generating function of is expressed as:

[0049]

[0050] Using the moment generating function, the false alarm probability and detection probability of OS-CFAR can be obtained as follows:

[0051]

[0052] It can be seen that the false alarm probability and detection probability of CA and OS-CFAR do not depend on μ, so they have the CFAR characteristics.

[0053] The detection window of the one-dimensional CFAR detector only slides in one direction, and the robustness of detection will decrease significantly when the background changes violently. Affected by the complex marine environment and airspace filtering algorithms, etc., the noise in the azimuth and range directions of sonar images changes violently. Therefore, in order to obtain better detection performance, a two-dimensional detection window is adopted in this embodiment. The cross detection window can greatly reduce the number of calculation reference units on the premise of taking into account the vertical range and horizontal azimuth noise of the target, and the calculation complexity is much lower than that of a rectangular detection window of the same size. Therefore, a cross detection window is adopted in this embodiment, and in order to reduce the computational load of the algorithm, a detection window with too large a size will not be used. When using a two-dimensional cross detection window to estimate the intensity of background clutter, the total number of reference units is R = 4n.

[0054] II. Segmentation-detection framework:

[0055] In sonar image data, the data points occupied by the target are extremely few. If two-dimensional CFAR detection is performed on each pixel point, a large amount of computing resources and processing time are required. Therefore, this embodiment proposes a segmentation-detection framework suitable for fast CFAR detection of sonar image targets, as Figure 2 shown. First, calculate the global segmentation threshold according to the different distribution characteristics of the two types of data, target and background, and quickly segment the target area; then calculate the local adaptive threshold in the segmented target area using a two-dimensional CFAR sliding detection window to achieve accurate detection of the target.

[0056] Assume that the size of a frame of sonar image data is M×N, and the amplitude level change range is {0, 1,..., L-1}. The frequency of amplitude i is f i , then the probability of the occurrence of this amplitude is:

[0057]

[0058] Taking the amplitude threshold k as the dividing line, the data with the amplitude level range of {0, 1,..., k} is classified into the background (I0) class, and the data with the amplitude level range of {k+1, k+2,..., L-1} is classified into the target (I1) class, so as to divide all the data into two classes, I0 and I1. Then the probabilities w0, w1 and the means μ0, μ1 of the two classes of data I0 and I1 are respectively:

[0059]

[0060] The overall mean of the data amplitude is:

[0061]

[0062] The between-class variance of the two classes of data is:

[0063] S B = w0(μ0 - μT ) 2 +w1(μ1 - μ T ) 2 (19)

[0064] Otsu's method uses the between-class variance as the evaluation function, representing the sum of the distances from two classes, I0 and I1, to the data center. Assume that when the amplitude threshold is k * , the between-class variance between I0 and I1 is the largest. At this time, the amplitude k * is the optimal amplitude segmentation threshold:

[0065]

[0066] Traverse the sonar data. If the amplitude of a certain data is greater than the optimal amplitude segmentation threshold, it is a target; otherwise, it is the background. Using Otsu's method can basically segment out all the target areas, remove most of the background, and significantly reduce the computational complexity of the subsequent detection algorithm. In addition, Otsu's segmentation also removes some obvious interferences that may trigger false alarms, so it helps to reduce false alarms of the detection algorithm.

[0067] III. SOW-CFAR Detection:

[0068] Based on the segmentation-detection framework, this embodiment proposes a SOW-CFAR detection method, as Figure 3 shown. First, use Otsu's method to segment the area where the target is located, and then perform two-dimensional ordered weighted CFAR detection based on a cross detection window that takes into account both azimuth and range noise.

[0069] Inspired by the OS class of CFAR detection algorithms, SOW-CFAR sorts the reference cells entering the sliding window from smallest to largest to obtain ordered samples:

[0070] x (1) ≤x (2) ≤…≤x (R) (21)

[0071] High-end ordered samples may lead to an overestimated background power, which is not conducive to detecting weak targets; low-end ordered samples may lead to an underestimated background power, resulting in an increase in false alarms. Therefore, different weighting coefficients should be assigned to the sorted reference cell samples, and the weighted sum of the ordered samples is used as the background power estimate:

[0072]

[0073] where ω i (i = 1, 2,..., R) are the weighting coefficients.

[0074] Although the reference cell variable x i(i ∈ {1, 2, …, R}) are statistically independent, but for the ordered statistics x obtained after sorting them (i) (i ∈ {1, 2, …, R}) are not statistically independent, and the probability density function of x cannot be simply convolved to find the probability density function of Z. Therefore, an auxiliary variable is introduced: (i) The probability density function of Z is obtained by convolution. Therefore, an auxiliary variable is introduced:

[0075] v i = (R + 1 - i)[x (i) - x (i-1) (1 ≤ i ≤ R, x (0) = 0) (23)

[0076] From the properties of the exponential distribution order statistics, it can be seen that v i (i ∈ {1, 2, …, R}) are independent of each other and follow an exponential distribution, and the probability density function is:

[0077]

[0078] Through simple algebraic operations, we have:

[0079]

[0080] Therefore, the moment generating function M Z (t) of Z can be expressed as:

[0081]

[0082] Using the moment generating function, the false alarm probability and the detection probability can be obtained as:

[0083]

[0084]

[0085] According to different scenarios, the weighting coefficient can be taken as rectangular or trapezoidal. When the weighting coefficient is taken as rectangular, the expression of the weighting coefficient is:

[0086]

[0087] where

[0088]

[0089] The rectangular weighting coefficient sets the weighting coefficients of the low-end and high-end ordered samples to zero, that is, the high-end and low-end ordered samples are removed, and the average of the remaining samples is used as an estimate of the background power. CA, OS, and censored-mean CFAR (CM-CFAR) can also be regarded as special cases of the rectangular weighting coefficient.

[0090]

[0091] When the weighting coefficient is trapezoidal, the weighting coefficient expression is:

[0092]

[0093] The trapezoidal weighting coefficient assigns small weighting coefficients to low-end and high-end samples instead of completely discarding them, and uses the weighted sum of all ordered samples to estimate the background power. Compared with the rectangular weighting coefficient, this method has higher balance and robustness, especially in clutter environments. The framework provides enhanced detection performance by adaptively combining all reference samples while mitigating the impact of extreme values.

[0094] The rectangular weighting coefficient enhances the robustness of the detection algorithm against multi-target interference and is especially suitable for short-range, high signal-to-noise ratio situations. However, in low signal-to-noise ratio environments, its performance will significantly degrade, often resulting in missed detections. In contrast, the trapezoidal weighting coefficient can ensure better target detection under different signal-to-noise ratio conditions but will increase false alarms. However, when the two-dimensional detection window using the trapezoidal weighting coefficient is combined with the segmentation algorithm, its false alarm rate can be effectively reduced. Therefore, by strategically integrating the rectangular and trapezoidal weighting coefficients, the detection system can achieve excellent overall performance, thus balancing anti-interference robustness and reliable target detection under different operating conditions. SOW-CFAR adopts a hybrid weighting coefficient strategy, using the rectangular weighting coefficient at close range to suppress strong sidelobe interference and the trapezoidal weighting coefficient at long range to reduce missed detections of low signal-to-noise ratio edge terrain echoes.

[0095] IV. Experiments:

[0096] 4.1 Lake experiment data:

[0097] In November 2023, a certain lake in northeastern China was selected as the experimental research area. The HT series multi-beam bathymeters developed by Harbin Engineering University were used for lake terrain data acquisition experiments to evaluate the detection performance of the proposed method. The HT series multi-beam bathymeters are based on Mill's cross-array sonar architecture, and the main technical parameters are: the center frequency is 200 kHz, the transmitted signal is a single-frequency rectangular pulse, and the receiving linear array is composed of 108 array elements.

[0098] The water body image of the multi-beam sonar bathymeter contains many terrain targets, the signal-to-noise ratio difference between the middle terrain and the edge terrain echoes is large, and there is also large sidelobe interference. Therefore, the performance of the detection algorithm can be evaluated higher. After amplifying, filtering, demodulating, and beamforming the original echo signal, the actual lake test data was obtained for verifying the proposed algorithm. The test data is as Figure 4 , Figure 5 shown, and the continuous strips in the figure are terrain targets.

[0099] 4.2. Analysis of test results:

[0100] 4.2.1. Performance analysis of the segmentation-detection framework:

[0101] To fully evaluate the effectiveness of the segmentation-detection framework proposed in this embodiment and the SOW-CFAR detection method, the most widely used CA and OS-CFAR in the CFAR detection algorithm are selected as the comparison algorithms. The segmentation-detection framework is applied to CA and OS to form the Segmented Cell Averaging (SCA) and Segmented Ordered Statistics (SOS) methods.

[0102] Use the multi-beam sonar sounding water body image 1( Figure 4 ) to test the detection performance of the four methods of CA, SCA, OS, and SOS, and analyze the performance of the segmentation-detection framework. The reference window length is set to n = 32, that is, R = 128, and the false alarm rate is set to 2%. The detection results of each method are as Figure 6 , Figure 7 , Figure 8 , Figure 9 shown.

[0103] Calculate the detection performance indicators of each method based on the detection results of the water body image data, including Precision, Accuracy, Recall, Missed Alarm Probability (MAR), False Alarm Rate (FAR), Elapsed Time, etc. The quantization indicators of the detection results of each method are shown in Table 1.

[0104] Table 1 Detection performance indicators of each algorithm for the multi-beam sounding water body image 1

[0105]

[0106] By comparing Figure 6 and Figure 7 and Figure 8 and Figure 9 the results shown in, it can be observed that compared with the traditional CA and OS-CFAR methods, the SCA and SOS-CFAR methods significantly reduce the number of false alarms, especially in the short-distance area lacking terrain targets. As shown in Table 1, after adopting this segmentation-detection framework, the FAR of the OS-CFAR method reduces the false alarm rate from 18.371% to 4.372%. This improvement can be attributed to the global segmentation step in this framework, which can effectively eliminate the obvious interference sources that may cause false alarms, thereby reducing the FAR of the detection method.

[0107] In addition, SCA-CFAR completed the detection process within 0.313 s, exceeding 3.936 s for CA, 10.687 s for OS, and 0.498 s for SOS-CFAR. This improvement in efficiency can be attributed to the global segmentation step of the framework, which eliminates most of the background regions and bypasses unnecessary local detection operations.

[0108] Therefore, the segmentation detection framework can effectively reduce the computational complexity and FAR of the CA and OS-CFAR detection methods in sonar image target detection.

[0109] 4.2.2. Performance analysis of the SOW-CFAR method:

[0110] The detection performance of three methods, namely CA, OS, and SOW, was tested using the multi-beam sonar sounding water body image 2( Figure 5 ). The reference window length was set to n = 32, i.e., R = 128, and the false alarm rate was set to 2%. The detection results of each method are shown in Figure 10 , Figure 11 , Figure 12 . The calculated results of the detection performance indicators of each algorithm are shown in Table 2.

[0111] Table 2 Detection performance indicators of each algorithm for the multi-beam sounding water body image 2

[0112]

[0113] By comparing the results given in Figure 10 , Figure 11 , Figure 12 , it can be concluded that compared with the traditional CA and OS-CFAR methods, the SOW-CFAR method achieves fewer false alarms in the short-distance area lacking terrain targets. As shown in Table 2, the accuracy of the SOW-CFAR method is 96.357% and the FAR is 3.643%, which is better than the CA and OS-CFAR methods. SOW-CFAR completed the detection process within 0.607 s, exceeding 4.016 s for CA and 10.855 s for OS-CFAR, but its computational complexity is higher than the other two methods. These improvements can be attributed to the proposed segmentation detection framework, which significantly removes the background regions, thus bypassing the computationally intensive detection operations for non-essential regions.

[0114] In the low signal-to-noise ratio far-edge terrain area, the SOW-CFAR method achieved a higher terrain target detection count than the traditional CA and OS-CFAR detectors. The MDR was 21.148%, which was better than the CA and OS-CFAR methods with MDR values of 57.443% and 46.038% respectively, as shown in Table 2. This improvement can be attributed to the trapezoidal weighting coefficient adopted by the SOW-CFAR method in these areas, which calculates a lower amplitude threshold, enabling the detection of more terrain targets.

[0115] Therefore, the SOW-CFAR method shows significant superiority over the CA and OS-CFAR methods in terms of computational efficiency, FAR, and MDR.

[0116] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be encompassed within the present invention.

[0117] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A fast constant false alarm detection method for sonar image targets based on segmented ordered weighting, characterized in that: Its detection method is as follows: Step 1: Adopt a segmentation-detection framework, combining global Otsu segmentation and local two-dimensional CFAR detection. Before local detection, use the global segmentation algorithm of Otsu method to reduce the computational complexity and false alarm rate in the subsequent detection process; Step 2: Adopt a segmentation-ordered weighted CFAR detection method. After segmenting the potential target area, use rectangular and trapezoidal ordered weighted coefficients for reference cell fusion in local two-dimensional CFAR detection to improve the adaptability of the detection algorithm to the complex detection background of sonar images, reduce the probability of missed detection, and enhance the detection performance.

2. The constant false alarm detection method for sonar image targets based on segmented ordered weighting according to claim 1, characterized in that: In the segmentation-detection framework in Step 1, first calculate the global segmentation threshold according to the different distribution characteristics of the two types of data of the target and the background, and quickly segment the target area; Then, calculate the local adaptive threshold using a two-dimensional CFAR sliding detection window in the segmented target area to achieve accurate detection of the target.

3. The constant false alarm detection method for sonar image targets based on segmented ordered weighting according to claim 1, characterized in that: In the segmentation-ordered weighted CFAR detection method, first use the Otsu method to segment the area where the target is located, and then perform two-dimensional ordered weighted CFAR detection based on a two-dimensional detection window that takes into account the noise in the azimuth and range directions.

Images