A method and system for processing two-dimensional imaging data of a ring scanning sonar

Abstract

The application relates to the field of image data processing, in particular to a ring-scan sonar two-dimensional imaging data processing method and system, which comprises the following steps: acquiring a ring-scan sonar image, performing superpixel segmentation on the ring-scan sonar image, taking any superpixel block as a target block, calculating the average noise pollution degree and the average echo signal intensity of the target block, taking the ratio of the average echo signal intensity to the average noise pollution degree as the comprehensive weight of the target block, taking the comprehensive weight of the target block as the comprehensive weight of each pixel point in the target block, traversing to obtain the comprehensive weight of each pixel point of each superpixel block, and constructing a weight map aligned with the ring-scan sonar image according to the comprehensive weights of all pixel points of all superpixel blocks; taking the ring-scan sonar image and the weight map as the double-channel input of a preset detection model to obtain a detection result. The application improves the target detection precision of the ring-scan sonar.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image data processing. More particularly, the present application relates to a method and system for processing two-dimensional imaging data of a ring scanning sonar. BACKGROUND

[0002] The ring scanning sonar generates underwater two-dimensional images by emitting sound waves and receiving echoes, and is widely used in underwater detection, target recognition and other scenarios. However, the underwater environment is complex, and the ring scanning sonar image is easily disturbed by noise (such as water scattering, equipment noise), and the target echo signal strength is superimposed with noise, resulting in low image signal-to-noise ratio and blurred details. Traditional image processing methods mostly rely on single image features or simple noise filtering, which is difficult to effectively distinguish targets from noise, thereby affecting the accuracy of subsequent target detection. SUMMARY

[0003] The main purpose of the embodiments of the present application is to provide a method and system for processing two-dimensional imaging data of a ring scanning sonar, aiming to improve the target detection accuracy of the ring scanning sonar.

[0004] To achieve the above-mentioned purpose, the embodiments of the first aspect of the present application provide a method for processing two-dimensional imaging data of a ring scanning sonar, the method comprising: acquiring a ring scanning sonar image, calculating the noise pollution degree of each pixel point in the ring scanning sonar image; performing superpixel segmentation on the ring scanning sonar image to obtain a plurality of superpixel blocks, taking any superpixel block as a target block, calculating the average noise pollution degree and the average echo signal strength of the target block, taking the ratio of the average echo signal strength to the average noise pollution degree as the comprehensive weight of the target block, taking the comprehensive weight of the target block as the comprehensive weight of each pixel point in the target block, traversing to obtain the comprehensive weight of each pixel point of each superpixel block, and constructing a weight map aligned with the ring scanning sonar image according to the comprehensive weight of all pixel points of all superpixel blocks; taking the ring scanning sonar image and the weight map as the dual-channel input of a preset detection model to obtain a detection result.

[0005] In some embodiments, calculating the noise pollution degree of each pixel point in the ring scanning sonar image comprises: taking any pixel point in the ring scanning sonar image as a target point, calculating the local noise intensity of the target point, performing a negative correlation mapping on the detection distance corresponding to each pixel point in the ring scanning sonar image using an exponential function, taking the difference between 1 and the mapping result of the target point as the distance weight of the target point, and taking the product of the normalized local noise intensity and the distance weight as the noise pollution degree of the target point; traversing to obtain the noise pollution degree of each pixel point in the ring scanning sonar image.

[0006] In some embodiments, the calculation of the local noise intensity of the target point comprises: constructing a sliding window centered on the target point, calculating the standard deviation and the mean of the gray scale of all pixel points in the sliding window, and taking the ratio of the standard deviation and the mean of the gray scale as the local noise intensity of the target point.

[0007] In some embodiments, the superpixel segmentation of the ring-scan sonar image comprises: performing a negative correlation mapping on the noise pollution degree of each pixel point in the ring-scan sonar image to obtain the reliability of each pixel point, and calculating the cumulative value of the reliability of all pixel points; taking any pixel point in the ring-scan sonar image as a reference point, taking the ratio of the reliability of the reference point and the cumulative value as the probability of the reference point being selected as an initial seed point, and traversing to obtain the probability of each pixel point in the ring-scan sonar image being selected as an initial seed point; and performing iteration based on the probability of each pixel point being selected as an initial seed point to segment the ring-scan sonar image.

[0008] In some embodiments, the training of the preset detection model comprises: obtaining a plurality of historical ring-scan sonar images and a weight map corresponding to each historical ring-scan sonar image; taking a historical ring-scan sonar image after labeling as an input of a first channel of an instance segmentation network, taking a weight map after labeling as an input of a second channel of the instance segmentation network, and training the instance segmentation network by using a multi-task loss function to obtain the preset detection model.

[0009] In some embodiments, the ring-scan sonar two-dimensional imaging data processing method further comprises: obtaining detection results of a plurality of continuous time instants, performing continuous multi-frame time sequence analysis on the plurality of detection results, and generating a detection report according to the analysis result.

[0010] In some embodiments, the continuous multi-frame time sequence analysis comprises: performing trajectory association analysis and motion consistency verification on the identified target object.

[0011] Embodiments of the second aspect of the present application propose a ring-scan sonar two-dimensional imaging data processing system, which comprises a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, the above-mentioned ring-scan sonar two-dimensional imaging data processing method is realized.

[0012] Advantages of the present application:

[0013] This invention uses the ratio of average echo signal strength to average noise contamination level as the comprehensive weight of the superpixel block: the greater the average echo signal strength (the stronger the target attribute, the more likely it is to be the target to be detected, such as an obstacle, rather than interference such as silt or water), the greater the comprehensive weight; the lower the average noise contamination level, the greater the comprehensive weight. The comprehensive weight of the superpixel block is mapped to each pixel in the superpixel block to construct a weight map aligned with the ring-scan sonar image. Then, the ring-scan sonar image and the weight are... Figure One The pre-defined detection model is used for target detection. Based on the weight map, the model can more accurately focus on high-weight regions (i.e., potential target areas with strong echoes and low noise) while weakening the influence of low-weight regions (such as noise interference areas or low-echo non-target areas). In this way, the pre-defined detection model can reduce the interference of noise and non-target information during feature learning and target recognition, and more efficiently capture the features of real targets. This improves the accuracy of target (such as obstacles) recognition in complex underwater environments, reduces false detection and false negative rates, and ultimately improves the target detection accuracy of the ring-scan sonar. Attached Figure Description

[0014] Figure 1 This is a flowchart of steps S1-S3 in a two-dimensional imaging data processing method for circumferential scanning sonar according to an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0016] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] Reference Figure 1 A method for processing two-dimensional imaging data from a ring-scan sonar includes steps S1-S3, as detailed below:

[0018] Step S1: Acquire the ring scan sonar image and calculate the noise pollution level of each pixel in the ring scan sonar image.

[0019] It should be further explained that each pixel in a ring-scan sonar image corresponds to a single acquisition point. The grayscale value (or color value) of a pixel directly reflects the echo signal intensity at that acquisition point: a higher grayscale value indicates a stronger ability of the object at that acquisition point to reflect sound waves (such as hard rocks, metal objects, etc.); a lower grayscale value indicates a weaker reflection ability (such as silt, water bodies, etc.). The horizontal and vertical coordinates of the pixel reflect the spatial information of that acquisition point, such as the azimuth angle and the distance between the acquisition point and the ring-scan sonar.

[0020] In some embodiments, calculating the noise pollution level of each pixel in the ring scan sonar image includes: taking any pixel in the ring scan sonar image as a target point, calculating the local noise intensity of the target point, using an exponential function to perform a negative correlation mapping on the detection distance corresponding to each pixel in the ring scan sonar image, taking the difference between 1 and the mapping result of the target point as the distance weight of the target point, and taking the product of the normalized local noise intensity and the distance weight as the noise pollution level of the target point; and iterating through and obtaining the noise pollution level of each pixel in the ring scan sonar image.

[0021] In some embodiments, calculating the local noise intensity of a target point includes: constructing a sliding window centered on the target point, calculating the gray standard deviation and gray mean of all pixels in the sliding window, and using the ratio of the gray standard deviation to the gray mean as the local noise intensity of the target point.

[0022] For example, the formula for calculating local noise intensity is as follows:

[0023]

[0024] In the formula middle, The first in the circumferential sonar image Local noise intensity of each pixel; For the first The standard deviation of grayscale values ​​of all pixels in the sliding window corresponding to each pixel; For the first The average grayscale value of all pixels in the sliding window corresponding to each pixel.

[0025] For the above formula It should be added that, The larger the value, the higher the value. The stronger the local noise intensity (the more dramatic the noise fluctuation) in the local region where the first pixel is located, the higher the local noise intensity (the more dramatic the noise fluctuation). The more susceptible a pixel is to local noise, the more likely the second pixel is to be affected. The lower the reliability of information from each pixel.

[0026] For example, the formula for calculating distance weight is as follows:

[0027]

[0028] In the formula middle, The first in the circumferential sonar image Distance weight of each pixel; It is an exponential function; For the first The detection distance corresponding to the pixel (circular scan sonar and the first pixel) (distance between the sampling points corresponding to each pixel) This is the decay constant, used to control the decay rate of the exponential function; This is the mapping result.

[0029] For the above formula It should be added that, The larger, the more The greater the detection distance of the first pixel, the better the detection result (the first pixel). The greater the noise affecting the grayscale value of a pixel, the more likely it is to be affected by noise; conversely, the less noise affects the grayscale value of the first pixel. The less susceptible the detection results of individual pixels are to noise, the more reliable they are. This aligns with the physical characteristics of low near-field signal attenuation and relatively minor noise interference.

[0030] For the above formula It should be added that, Depending on the specific scenario, for example, based on industry experience, 455kHz sonar corresponds to... It can be 20 meters.

[0031] For example, the formula for calculating the degree of noise pollution is as follows:

[0032]

[0033] In the formula middle, The first in the circumferential sonar image The degree of noise contamination per pixel For normalized , For the first Local noise intensity of each pixel For the first Distance weights for each pixel.

[0034] For the above formula It should be added that, The larger the value, the higher the value. The more severe the overall contamination of a pixel, the lower its information reliability. This invention calculates the degree of noise contamination by combining local noise intensity with distance weights. It considers both the noise distribution around the pixel and the characteristic of sonar detection that "the farther the distance, the greater the noise impact," making the noise assessment more in line with the actual scenario and providing a reliable basis for subsequent weight calculations.

[0035] Step S2: Perform superpixel segmentation on the circumferential sonar image to obtain several superpixel blocks. Take any superpixel block as the target block, calculate the average noise pollution level and average echo signal intensity of the target block, and use the ratio of the average echo signal intensity to the average noise pollution level as the comprehensive weight of the target block. Use the comprehensive weight of the target block as the comprehensive weight of each pixel in the target block, traverse to obtain the comprehensive weight of each pixel in each superpixel block, and construct a weight map aligned with the circumferential sonar image based on the comprehensive weights of all pixels in all superpixel blocks.

[0036] It should be further explained that the circumferential scan sonar image is aligned with the weight map, meaning that a pixel in the circumferential scan sonar image corresponds to a weight in the same position in the weight map. For example, the pixel in the 2nd row and 3rd column of the circumferential scan sonar image corresponds to a weight in the 2nd row and 3rd column of the weight map.

[0037] It should be added that the purpose of superpixel segmentation is to divide an image into compact groups of pixels with similar features, replacing a large number of original pixels with a small number of superpixels, thereby reducing the complexity of subsequent processing.

[0038] In some embodiments, superpixel segmentation of a ring scan sonar image includes: performing a negative correlation mapping on the noise contamination level of each pixel in the ring scan sonar image to obtain the reliability of each pixel, and calculating the cumulative value of the reliability of all pixels; taking any pixel in the ring scan sonar image as a reference point, using the ratio of the reliability of the reference point to the cumulative value as the probability that the reference point is selected as an initial seed point, and iterating to obtain the probability that each pixel in the ring scan sonar image is selected as an initial seed point; and iterating based on the probability that each pixel is selected as an initial seed point to segment the ring scan sonar image.

[0039] For example, the probability of being selected as the initial seed point is calculated using the following formula:

[0040]

[0041] In the formula middle, The first in the circumferential sonar image The probability of a pixel being selected as the initial seed point. For the first The degree of noise contamination per pixel For the first Reliability of each pixel This is the cumulative value of the reliability of all pixels.

[0042] For the above formula It should be added that, The smaller, The larger the value, the greater the probability that a pixel with less noise contamination will be selected as an initial seed point. This invention makes pixels with less noise and higher reliability more likely to become initial seed points, improving the accuracy of superpixel segmentation and making the segmented superpixel blocks fit the real target boundary more closely. The iterative segmentation process after selecting the initial seed point is a known existing technique and will not be described in detail in this invention.

[0043] It should be further explained that after segmenting the ring scan sonar image, multiple superpixel blocks are obtained. For each superpixel block, the average noise level of all pixels in the superpixel block is taken as the average noise level of the superpixel block, and the average echo signal intensity of all pixels in the superpixel block is taken as the average echo signal intensity of the superpixel block.

[0044] For example, the formula for calculating the overall weight of a superpixel block is as follows:

[0045]

[0046] In the formula middle, For the first The overall weight of each superpixel block, For the first Average echo signal strength of each superpixel block For the first The average noise pollution level of each superpixel block.

[0047] For the above formula It should be noted that the higher the average echo signal strength and the lower the average noise pollution level, the greater the probability that the superpixel block is a potential underwater obstacle area, and the higher the corresponding comprehensive weight should be. When using the model for detection in the future, the model will focus more on this area.

[0048] For the above formula It should be added that the first The combined weight of each superpixel block is , No. The overall weight of each pixel in each superpixel block is also... .

[0049] Step S3: Use the ring scan sonar image and weight map as dual-channel inputs to the preset detection model to obtain the detection results.

[0050] In some embodiments, training the preset detection model includes: acquiring multiple historical ring scan sonar images and a weight map corresponding to each historical ring scan sonar image; using the labeled historical ring scan sonar images as input to the first channel of the instance segmentation network, using the corresponding labeled weight map as input to the second channel of the instance segmentation network, and training the instance segmentation network using a multi-task loss function, thereby obtaining the preset detection model after training is completed.

[0051] It should be noted that the labels include information such as the pixel's category. The multi-task loss function includes a classification loss function and a mask segmentation loss function. During training, the instance segmentation network preferentially learns the above formulas. The pixels with a relatively large overall weight.

[0052] It should be noted that the detection results are the output of the preset detection model, including the target bounding box, category, confidence score, etc.

[0053] In some embodiments, the ring-scan sonar two-dimensional imaging data processing method further includes: acquiring multiple detection results at different consecutive times, performing continuous multi-frame temporal analysis on the multiple detection results, and generating a detection report based on the analysis results.

[0054] In some embodiments, continuous multi-frame temporal analysis includes: performing trajectory association analysis and motion consistency verification on the identified target objects.

[0055] It should be noted that trajectory correlation analysis and motion consistency verification can be achieved using existing well-known techniques, which will not be described in detail here.

[0056] It should be noted that the test report includes the precise outline, category, and confidence level of the target object.

[0057] The present invention also provides a two-dimensional imaging data processing system for circumferential scanning sonar. The system includes a processor and a memory, the memory storing computer program instructions. When the computer program instructions are executed by the processor, they implement a two-dimensional imaging data processing method for circumferential scanning sonar according to the first aspect of the present invention. The system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface; their configuration and functions are known in the art and will not be described further here.

[0058] It should be noted that the preferred embodiments of this application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of this application. For those skilled in the art, various modifications and improvements can be made without departing from the concept of the invention, and these all fall within the protection scope of the invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for processing two-dimensional imaging data from a circular scanning sonar system, characterized in that, include: Acquire a ring scan sonar image and calculate the noise contamination level of each pixel in the ring scan sonar image; The circumferential sonar image is segmented into several superpixel blocks. Any superpixel block is taken as a target block. The average noise pollution level and average echo signal intensity of the target block are calculated. The ratio of the average echo signal intensity to the average noise pollution level is taken as the comprehensive weight of the target block. The comprehensive weight of the target block is taken as the comprehensive weight of each pixel in the target block. The comprehensive weight of each pixel in each superpixel block is obtained by traversing. A weight map aligned with the circumferential sonar image is constructed based on the comprehensive weights of all pixels in all superpixel blocks. The detection results are obtained by using the circumferential sonar image and the weight map as dual-channel inputs to a preset detection model.

2. The method for processing two-dimensional imaging data of a circular scanning sonar according to claim 1, characterized in that, Calculating the noise contamination level of each pixel in the circumferential sonar image includes: Taking any pixel in the ring scan sonar image as the target point, the local noise intensity of the target point is calculated. An exponential function is used to perform a negative correlation mapping on the detection distance corresponding to each pixel in the ring scan sonar image. The difference between 1 and the mapping result of the target point is taken as the distance weight of the target point. The product of the normalized local noise intensity and the distance weight is taken as the noise pollution degree of the target point. The noise contamination level of each pixel in the circumferential sonar image is obtained by iterating through the images.

3. The method for processing two-dimensional imaging data of a circular scanning sonar according to claim 2, characterized in that, Calculating the local noise intensity at the target point includes: A sliding window is constructed with the target point as the center. The gray standard deviation and gray mean of all pixels in the sliding window are calculated. The ratio of the gray standard deviation to the gray mean is used as the local noise intensity of the target point.

4. The method for processing two-dimensional imaging data of a circular scanning sonar according to claim 1, characterized in that, Superpixel segmentation of the circumferential sonar image includes: The noise contamination level of each pixel in the ring scan sonar image is negatively correlated to obtain the reliability of each pixel, and the cumulative value of the reliability of all pixels is calculated. Using any pixel in the circumferential sonar image as a reference point, the ratio of the reliability of the reference point to the accumulated value is used as the probability that the reference point is selected as the initial seed point. The probability that each pixel in the circumferential sonar image is selected as the initial seed point is obtained by traversing the image. The circumferential sonar image is segmented by iterating based on the probability that each pixel is selected as an initial seed point.

5. The method for processing two-dimensional imaging data of a circumferential sonar according to claim 1, characterized in that, The training of the preset detection model includes: Acquire multiple historical ring scan sonar images and the weight map corresponding to each historical ring scan sonar image; The historical ring-scan sonar images with labels are used as the input to the first channel of the instance segmentation network, and the weight maps with corresponding labels are used as the input to the second channel of the instance segmentation network. The instance segmentation network is trained using a multi-task loss function, and the preset detection model is obtained after training.

6. The method for processing two-dimensional imaging data of a circular scanning sonar according to claim 1, characterized in that, The ring-scan sonar two-dimensional imaging data processing method further includes: The system acquires multiple detection results at different consecutive time points, performs continuous multi-frame temporal analysis on these results, and generates a detection report based on the analysis results.

7. The method for processing two-dimensional imaging data of a circular scanning sonar according to claim 6, characterized in that, The continuous multi-frame timing analysis includes: Trajectory association analysis and motion consistency verification are performed on the identified target objects.

8. A two-dimensional imaging data processing system for circumferential scanning sonar, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a two-dimensional imaging data processing method for circumferential scanning sonar according to any one of claims 1-7.

Images