Riverbed sediment enhancement method and system fusing confidence learning and superpixel segmentation

Abstract

The invention belongs to the technical field of riverbed sediment exploration, and discloses a riverbed sediment enhancement method and system fusing confidence learning and superpixel segmentation. According to the method, a superpixel segmentation method is used to divide a multi-beam backscattering intensity image; carrying out sample expansion on the pixels after the super-pixel segmentation; and sample enhancement is carried out by using confidence learning. According to the method, on the basis of on-site riverbed substrate sampling points, a superpixel segmentation method is used for segmenting a multi-beam backscattering intensity image into different superpixel areas, the center of a superpixel block serves as a representation of a superpixel, noise in an expansion data set is removed through confidence learning, and riverbed substrate sample enhancement is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of riverbed sediment exploration, and in particular relates to a riverbed bottom enhancement method and system integrating confidence learning and superpixel segmentation. Background Art

[0002] As a key geographical element that carries the benthic ecology of the river basin and connects rivers, lakes, deltas and estuaries, rapid and effective detection and identification of underwater sediments in river basins are important for studying the distribution and migration process of river-dwelling organisms, calculating the sediment-carrying capacity of water flow, bedload transport rate and analysis, predicting the flow force of river evolution, and predicting the evolution trend of rivers. As the sink and source of various substances in the basin, riverbed sediments are of great significance in river water environment protection and even in the improvement of basin ecological environment and restoration of ecosystem functions.

[0003] At present, riverbed sediment exploration technologies mainly include traditional field sampling technology and acoustic remote sensing detection methods based on acoustic media. The multi-beam echo sounder system can obtain continuous depth and backscatter intensity data over a large range. By combining traditional field sampling with acoustic detection, it can characterize the spatial distribution and characteristics of riverbed sediments over a large range, and has become a basic tool for measuring the main biological and non-biological characteristics necessary for the habitat.

[0004] Traditional underwater in-situ sampling technology uses direct sampling techniques such as box sampling, tube sampling, trawl sampling, underwater photography, grab sampling and artificial diving sampling to collect bottom samples. However, these technologies usually have problems such as complex operation, long time consumption and sampling difficulties. Limited by the bottom sampling method, the scarcity of bottom sampling has become a key problem hindering the further development of bottom classification. With the maturity of underwater detection technology, the resolution of multi-beam sonar systems has been continuously improved, resulting in a gradual gap between the number of underwater field sampling samples and the volume of acoustic data. The resulting model overfitting problem will affect the final classification accuracy. In response to the above problems, the current mainstream solutions are: first, the underwater acoustic image is divided into multiple small-scale images, and the sediment type of each small-scale image is determined by direct sampling analysis, so as to obtain an expanded sample of sediment for classification; second, the bottom type of a fixed neighborhood of the sampling point is obtained based on the distribution of surface sediment sample data and historical sediment distribution, so that the number of samples meets the needs of building a supervised classification model for the seabed bottom. The above two sample enhancement methods are both based on the principle of expanding limited seabed sediment sampling points: expanding regularly from known sample points outward. This method is suitable for small-scale sea areas with fewer sediment types, and can construct a higher-precision classification model.

[0005] In fact, the diverse forms of riverbed benthic environments lead to spatial irregularity in the variation of the sediments. The prominent features are the fuzziness of the sediment categories and the multi-scale span of the sediment boundaries, which conflict with the fixed neighborhood sample expansion method. At present, the concept of superpixel segmentation has been introduced in the sample expansion of seabed sediment classification, such as superpixel segmentation of multi-beam backscatter intensity images, and screening of samples based on the characteristic distance between the original sampling points and other sampling points in the expansion area; active learning based on voting entropy is used to screen superpixel expansion samples to solve the problem of small and unstable expansion samples. However, affected by the changes in river hydrodynamic conditions and the fluctuations of the riverbed base, sediments are constantly eroded and deposited, and the riverbed alluvial layer has a high degree of spatial heterogeneity and temporal variability in physical properties and biogeochemical processes. The above methods often lead to the possibility of multiple categories mixed in the segmented pixel blocks, especially the interference of non-attribute noise signals, which brings difficulties to classification. Summary of the invention

[0006] In order to overcome the problems existing in the related art, the disclosed embodiments of the present invention provide a riverbed sediment enhancement method and system integrating confidence learning and superpixel segmentation.

[0007] The technical solution is as follows: The riverbed bottom quality enhancement method integrating confidence learning and superpixel segmentation includes the following steps: S1, segmentation of multibeam backscatter intensity image using superpixel segmentation method; S2, sample expansion of pixels after superpixel segmentation; S3, sample enhancement using confidence learning.

[0008] In step S1, the multi-beam backscatter intensity image is segmented using a superpixel segmentation method, including: S101, initialize seed points and distribute them evenly in the backscattered image seed points, each superpixel size is ; S102, optimize the seed point, calculate the gradient of all pixels in the neighborhood of the initial seed point, and move the seed point to the gradient in the neighborhood. The gradient represents the direction and amplitude of the fastest change of pixel values ​​in the image. For grayscale images, the pixel value range is [0,255], and the theoretical range of the gradient amplitude is ; S103, calculating distance metrics, including color distance and spatial distance.

[0009] In step S103, the distance metric is calculated, and the formula is: (1) (2) (3) (4) In the formula, is the color distance, is the spatial distance, is the distance metric, is the maximum spatial distance within a cluster, is the brightness value of the superpixel seed point, Pixel The brightness of the multibeam backscatter intensity map is the grayscale. is the horizontal coordinate of the superpixel seed point, is the horizontal coordinate of the pixel point, is the ordinate of the superpixel seed, is the vertical coordinate of the pixel, is the maximum color distance, is the total number of pixels in the image, is the number of pre-segmented superpixels.

[0010] In step S2, sample expansion includes: when there is only one label in the superpixel block, labeling the entire superpixel block according to the label; when there are multiple labels, using the majority voting method to select the label mode of the labeled samples in the pixel block as the label of the superpixel block, and if the majority voting is not satisfied, re-segmenting the superpixel; definition The value of satisfies the following formula: ; in, The parameters set for the first split, if it does not meet the majority vote, let , re-dividing until the majority vote is met.

[0011] In step S3, sample enhancement is performed using confidence learning, including: First, feature extraction is performed on multi-beam water depth and backscatter intensity data; Secondly, combined with the on-site bottom sampling points, confidence learning is used to remove noise labels from the samples expanded by superpixel segmentation.

[0012] Further, remove the noise labels, including: S301, estimate the joint distribution of noise labels and true labels ; S302, data cleaning, set the error label to ,in, is an error label, For categories, To estimate the probability, is a noisy label, For categories, For a given sample, is the model parameter; the noise label is estimated through the off-diagonal elements, the noise samples in the expanded samples based on superpixel segmentation are found, and the noise samples are cleaned; S303, retraining the model, putting the noise-removed samples into the bottom sediment classification model for retraining to obtain a new bottom sediment classification result.

[0013] In step S301, the joint distribution of the noise label and the true label is estimated ,include: The estimated probability of cross validation is expressed as: (5) In the formula, Characterized by is the number of folds for data cross validation, For the Folding model pair No. Output probability of category prediction; In the count matrix, the diagonal captures the correct labels and the off-diagonal captures the asymmetric incorrect labels, expressed as: (6) (7) (8) In the formula, is the count matrix; is the noisy label of the sample, that is, the label provided by the dataset The value of is the true label of the sample, For the dataset In the A subset of samples; is the average confidence threshold for each category, For the dataset In the A subset of samples, For observation category, actually A collection of categories, is a subset of samples with noise, is an observable noisy label; is the true label without noise; For the model in parameters Next, sample Belongs to the tag The probability of The estimation formula for the joint distribution is: (9) In the formula, To be marked Among the samples, the actual category The number of samples, The noisy label is The size of the sample set, For the noisy label In the sample of , the true label is The probability of A collection of categories.

[0014] Another object of the present invention is to provide a riverbed bottom sample enhancement system integrating confidence learning and superpixel segmentation, the system implements the riverbed bottom sample enhancement method integrating confidence learning and superpixel segmentation, and the system comprises: A superpixel segmentation module, for segmenting the multi-beam backscatter intensity image using a superpixel segmentation method; A sample expansion module, used to expand the samples of pixels after superpixel segmentation; Confidence learning module, used to perform sample enhancement using confidence learning.

[0015] Furthermore, the system is mounted on a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, and when the processor executes the computer program, the functions of the riverbed sediment sample enhancement system integrating confidence learning and superpixel segmentation are implemented.

[0016] Furthermore, the system is mounted on a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can realize the functions of the riverbed sediment sample enhancement system that integrates confidence learning and superpixel segmentation.

[0017] Combining all the above-mentioned technical solutions, the beneficial effects of the present invention are as follows: the present invention provides a riverbed sediment classification sample enhancement method that integrates superpixel segmentation and confidence learning. Based on the on-site riverbed sediment sampling points, the multi-beam backscatter intensity image is segmented into different superpixel areas using a superpixel segmentation method, and the center of the superpixel block is used as the representative of the superpixel. Confidence learning is used to remove noise in the expanded data set to achieve riverbed sediment sample enhancement.

[0018] The present invention reduces the cost of field sampling and improves the classification performance of the model. With the same amount of training data, the classification accuracy is improved, and a more accurate classification result of riverbed sediments is obtained. In precise bottom mapping, the present invention supports engineering decisions such as river ecological restoration and waterway dredging; in environmental monitoring optimization, it provides a high-precision data basis for sediment migration and pollutant diffusion research.

[0019] Existing methods at home and abroad mostly rely on field sampling to obtain the bottom type of a fixed neighborhood of the sampling point or segment the underwater acoustic image into multiple small-scale images, and determine the sediment type of each small-scale image through direct sampling analysis, lacking an automated and highly robust sample enhancement solution. This invention combines superpixel segmentation and confidence learning for the first time: realizing an end-to-end automated process from multi-beam backscatter intensity image segmentation to sample noise cleaning based on field sampling samples.

[0020] The sampling points of riverbed sediment are sparse and costly, and the traditional field sampling method is inefficient and costly. The present invention generates a large number of candidate samples based on superpixel segmentation to cover areas that are not directly sampled, thereby completing sample expansion. Confidence learning is used to extract high-quality data from candidate samples and perform sample enhancement, thereby effectively alleviating the problem of overfitting of small samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure; Figure 1 It is a schematic diagram of a riverbed bottom quality enhancement method integrating confidence learning and superpixel segmentation provided in an embodiment of the present invention; Figure 2 It is a flow chart of a riverbed bottom quality enhancement method integrating confidence learning and superpixel segmentation provided by an embodiment of the present invention; Figure 3 is a superpixel segmentation sample expansion flow chart provided by an embodiment of the present invention; Figure 4 It is an effect diagram containing only one type of label in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 5 This is an effect diagram of two types of labels with the same number in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 6 This is a diagram showing the effect of two types of labels with different numbers in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 7 It is a type of label expansion result graph in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 8It is a result diagram of expansion of two types of labels of the same number in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Fig. 9 It is a result diagram of label expansion of two types with different numbers in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Fig.10 is a sample enhancement result diagram using neighborhood window expansion provided by an embodiment of the present invention; Fig.11 is a sample enhancement result diagram based on superpixel segmentation and confidence learning in the present invention provided by an embodiment of the present invention; Fig.12 It is a depth measurement value map of a multi-beam measurement area in a certain region provided by an embodiment of the present invention; the horizontal and vertical coordinates are projection coordinates; Fig.13 is a backscatter intensity value diagram of a multi-beam measurement area in a certain region provided by an embodiment of the present invention; wherein, Fig.13 The geographical locations and categories of the 18 on-site sampling points are marked in the figure, and the horizontal and vertical coordinates are projection coordinates; Fig.14 It is a bottom sediment classification result diagram without using the sample enhancement method of the present invention in the comparison of bottom sediment classification results using the support vector machine classification algorithm provided by the embodiment of the present invention; the horizontal and vertical coordinates are projection coordinates; Fig.15 It is a bottom sediment classification result diagram using the sample enhancement method of the present invention in comparison with the bottom sediment classification result using the support vector machine classification algorithm provided by an embodiment of the present invention, and the horizontal and vertical axes are projection coordinates. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0023] The innovation of the present invention lies in that the sample enhancement method proposed in the present invention combines superpixel segmentation with confidence learning, constructs a sediment data set based on field sampling points, eliminates the noise in the sediment data set, makes full use of the information of the field sampling points, effectively mines the information in the unlabeled samples, and has significantly improved the accuracy of sample selection, thereby improving the accuracy of bottom sediment classification.

[0024] Embodiment 1, as Figure 1As shown, the riverbed sediment enhancement method integrating confidence learning and superpixel segmentation provided in an embodiment of the present invention combines confidence learning to remove noise from the sediment sampling points expanded by superpixel segmentation, performs effective sample enhancement on the original sampling points, and puts the noise-removed samples into the classification model for retraining to obtain more accurate sediment classification results, thereby providing a new solution for riverbed sediment classification sample enhancement.

[0025] like Figure 2 As shown, specifically including: S1, segmentation of multibeam backscatter intensity image using superpixel segmentation method; Superpixel segmentation uses color space information and spatial information to perform local clustering of multi-beam backscatter intensity image pixels to generate compact and approximately uniform superpixels, which can better express the texture information between objects. At the same time, when there is a lot of segmentation data, superpixel segmentation can obtain homogeneous areas with high purity. Specifically include: S101, initialize seed points and distribute them evenly in the backscattered image seed points, each superpixel size is ; S102, optimize the seed point, calculate the gradient of all pixels in the neighborhood of the initial seed point, and move the seed point to the gradient in the neighborhood. The gradient represents the direction and amplitude of the fastest change of pixel values ​​in the image. For grayscale images, the pixel value range is [0,255], and the theoretical range of the gradient amplitude is ; S103, calculating distance metrics, including color distance and spatial distance.

[0026] Calculate the distance metric, the formula is: (1) (2) (3) (4) In the formula, is the color distance, is the spatial distance, is the distance metric, is the maximum spatial distance within a cluster, is the brightness value of the superpixel seed point, Pixel The brightness of the multibeam backscatter intensity map is the grayscale. is the horizontal coordinate of the superpixel seed point, is the horizontal coordinate of the pixel point, is the ordinate of the superpixel seed, is the vertical coordinate of the pixel, is the maximum color distance, is the total number of pixels in the image, is the number of pre-segmented superpixels.

[0027] S2, sample expansion of pixels after superpixel segmentation; Combined with the location of the on-site sampling points, the on-site sampling points are marked in the multi-beam backscatter intensity map to obtain the association between samples and superpixel blocks. Category labels are set for samples of each category to facilitate sample expansion and enhancement.

[0028] The sample expansion strategy is as follows: when there is only one label in a superpixel block, the entire superpixel block is labeled according to the label; when there are multiple labels, the majority voting method is used to select the label mode of the labeled samples in the pixel block as the label of the superpixel block. If the majority vote is not satisfied, the superpixel is re-segmented.

[0029] definition The value of satisfies the following formula: ; in, The parameters set for the first split, if it does not meet the majority vote, let , re-dividing until the majority voting method is met. Figure 3 As shown, a flow chart of the sample expansion strategy is presented.

[0030] like Figure 4 It is an effect diagram containing only one type of label in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 5 This is an effect diagram of two types of labels with the same number in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 6 This is a diagram showing the effect of two types of labels with different numbers in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 7 It is a type of label expansion result graph in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Figure 8 It is a result diagram of expansion of two types of labels of the same number in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; Fig. 9 It is a result diagram of two types of label expansion with different numbers in the superpixel segmentation sample expansion strategy provided by an embodiment of the present invention; an example of superpixel expansion is given, in which a multi-beam backscatter intensity image is segmented using a superpixel segmentation method, and the scale number n is 17. Figure 4 and Figure 6 The expanded result after satisfying the majority vote is given. Figure 5 If the majority vote is not satisfied, the scale number n is adjusted to 18, and the adjacent sampling points fall within a superpixel block.

[0031] S3, sample enhancement using confidence learning.

[0032] Before performing sample enhancement on the expanded samples based on superpixel segmentation, firstly, feature extraction is performed on the multi-beam water depth and backscatter intensity data. Secondly, combined with the on-site bottom sampling points, the noise labels of the samples expanded by superpixel segmentation are removed through confidence learning. The steps for removing noise labels are as follows.

[0033] S301, estimate the joint distribution of noise labels and true labels ; In order to estimate the joint distribution, cross-validation is used to estimate the superpixel expanded labels to obtain the category probability of each sample ; Calculate the count matrix based on this probability and perform calibration , and finally estimate the joint distribution .

[0034] The estimated probability of cross validation is expressed as: (5) In the formula, Characterized by is the number of folds for data cross validation, For the Folding model pair No. Output probability of category prediction; In the count matrix, its diagonal captures the correct labels and the off-diagonal captures the asymmetric incorrect labels, which is calculated as follows: (6) (7) (8) In the formula, is the count matrix; is the noisy label of the sample, that is, the label provided by the dataset The value of is the true label of the sample, For the dataset In the A subset of samples; is the average confidence threshold for each category, For the dataset In the A subset of samples, For observation category, actually A collection of categories, is a subset of samples with noise, is an observable noisy label; is the true label without noise; For the model in parameters Next, sample Belongs to the tag The probability of The estimation formula for the joint distribution is: (9) In the formula, To be marked Among the samples, the actual category The number of samples, The noisy label is The size of the sample set, For the noisy label In the sample of , the true label is The probability of A collection of categories.

[0035] S302, data cleaning, set the error label to ,in, is an error label, For categories, To estimate the probability, is a noisy label, For categories, For a given sample, is the model parameter; the noise label is estimated through the off-diagonal elements, the noise samples in the expanded samples based on superpixel segmentation are found, and the noise samples are cleaned; S303, retraining the model, putting the noise-removed samples into the bottom sediment classification model for retraining to obtain a new bottom sediment classification result.

[0036] Fig.10 The sample enhancement result graph using neighborhood window expansion is shown. Fig.11 The sample enhancement result diagram based on superpixel segmentation and confidence learning in the present invention is shown. Fig.11 It can be seen that the present invention can better combine the texture features of the image than the neighborhood window expansion.

[0037] It can be seen from the above embodiments that the role of the superpixel segmentation algorithm of the present invention is to segment the multi-beam backscatter intensity image, and after segmentation, the sample is expanded in combination with the position of the on-site bottom sampling sample points; the role of confidence learning is to remove the noise sample points from the samples after superpixel segmentation and expansion. The present invention combines confidence learning and superpixel segmentation to segment and obtain homogeneous areas with high purity, and can expand and denoise the rare sample points.

[0038] The present invention divides the multi-beam backscatter intensity image based on superpixel segmentation, expands the segmentation results around the on-site substrate sampling points, and realizes the effective expansion of riverbed substrate samples; the noise samples in the samples expanded by superpixel segmentation are determined through confidence learning, and the noise samples are cleaned up, and the noise samples are combined with the noise-removed samples to be substituted into the classification model for retraining, so as to obtain more accurate substrate classification results, providing important technical support for riverbed substrate classification.

[0039] Embodiment 2, the embodiment of the present invention provides a riverbed sediment sample enhancement system integrating confidence learning and superpixel segmentation, comprising: A superpixel segmentation module, for segmenting the multi-beam backscatter intensity image using a superpixel segmentation method; A sample expansion module, used to expand the samples of pixels after superpixel segmentation; Confidence learning module, used to perform sample enhancement using confidence learning.

[0040] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0041] The present invention takes the multi-beam data collected on-site in a river basin in a certain region of a province as an example to compare the effects. Figure 12-13 The results of multi-beam bathymetry and backscattering intensity in an experimental area in a certain region are shown. First, feature extraction is performed on the multi-beam bathymetry and backscattering intensity data. Secondly, the multi-beam backscattering intensity image is preprocessed by a superpixel segmentation algorithm, and the backscattering image is segmented into several irregular superpixel regions. If there are field samples, sample expansion is performed; then, confidence learning is used to purify the expanded samples to remove noise samples. Finally, the samples are put into the classification model to generate classification results. Figure 14-15 The results of riverbed sediment classification using the support vector machine classification algorithm are shown. Fig.14 The total accuracy of the bottom classification in the experiment was 78.6%, and the Kappa coefficient was 0.704. Fig.15 The total accuracy of the bottom sediment classification is 79.8%, and the Kappa coefficient is 0.722. After using the sample enhancement method proposed in the present invention, the accuracy of the riverbed bottom sediment classification result is higher.

[0042] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A riverbed enhancement method integrating confidence learning and superpixel segmentation, characterized in that: The method comprises the following steps: S1, segmentation of multibeam backscatter intensity image using superpixel segmentation method; S2, sample expansion of pixels after superpixel segmentation; S3, sample enhancement using confidence learning.

2. The riverbed enhancement method integrating confidence learning and superpixel segmentation according to claim 1 is characterized in that: In step S1, the multi-beam backscatter intensity image is segmented using a superpixel segmentation method, including: S101, initialize seed points and distribute them evenly in the backscattered image seed points, each superpixel size is ; S102, optimize the seed point, calculate the gradient of all pixels in the neighborhood of the initial seed point, and move the seed point to the gradient in the neighborhood. The gradient represents the direction and amplitude of the fastest change of pixel values ​​in the image. For grayscale images, the pixel value range is [0,255], and the theoretical range of the gradient amplitude is ; S103, calculating distance metrics, including color distance and spatial distance.

3. The riverbed enhancement method integrating confidence learning and superpixel segmentation according to claim 2 is characterized in that: In step S103, the distance metric is calculated, and the formula is: （1） （2） （3） （4） In the formula, is the color distance, is the spatial distance, is the distance metric, is the maximum spatial distance within a cluster, is the brightness value of the superpixel seed point, Pixel The brightness of the multibeam backscatter intensity map is the grayscale. is the horizontal coordinate of the superpixel seed point, is the horizontal coordinate of the pixel point, is the ordinate of the superpixel seed, is the vertical coordinate of the pixel, is the maximum color distance, is the total number of pixels in the image, is the number of pre-segmented superpixels.

4. The riverbed enhancement method integrating confidence learning and superpixel segmentation according to claim 3 is characterized in that: In step S2, sample expansion includes: when there is only one label in the superpixel block, labeling the entire superpixel block according to the label; when there are multiple labels, using the majority voting method to select the label mode of the labeled samples in the pixel block as the label of the superpixel block, and if the majority voting is not satisfied, re-segmenting the superpixel; definition The value of satisfies the following formula: ； in, The parameters set for the first split, if it does not meet the majority vote, let , re-dividing until the majority vote is met.

5. The riverbed enhancement method integrating confidence learning and superpixel segmentation according to claim 3 is characterized in that: In step S3, sample enhancement is performed using confidence learning, including: First, feature extraction is performed on multi-beam water depth and backscatter intensity data; Secondly, combined with the on-site bottom sampling points, confidence learning is used to remove noise labels from the samples expanded by superpixel segmentation.

6. The riverbed enhancement method integrating confidence learning and superpixel segmentation according to claim 5 is characterized in that: Remove noisy labels, including: S301, estimate the joint distribution of noise labels and true labels ; S302, data cleaning, set the error label to ,in, is an error label, For categories, To estimate the probability, is a noisy label, For a given sample, is the model parameter; the noise label is estimated through the off-diagonal elements, the noise samples in the expanded samples based on superpixel segmentation are found, and the noise samples are cleaned; S303, retraining the model, putting the noise-removed samples into the bottom sediment classification model for retraining to obtain a new bottom sediment classification result.

7. The riverbed enhancement method integrating confidence learning and superpixel segmentation according to claim 6 is characterized in that: In step S301, the joint distribution of the noise label and the true label is estimated ,include: The estimated probability of cross validation is expressed as: （5） In the formula, Characterized by is the number of folds for data cross validation, For the Folding model pair No. Output probability of category prediction; In the count matrix, the diagonal captures the correct labels and the off-diagonal captures the asymmetric incorrect labels, expressed as: （6） （7） （8） In the formula, is the count matrix; is the noisy label of the sample, that is, the label provided by the dataset The value of is the true label of the sample, For the dataset In the A subset of samples; is the average confidence threshold for each category, is the subset of samples labeled as category 𝑗 in the dataset 𝑋, For observation category, actually A collection of categories, is a subset of samples with noise, is an observable noisy label; is the true label without noise; For the model in parameters Next, sample Belongs to the tag The probability of The estimation formula for the joint distribution is: （9） In the formula, To be marked Among the samples, the actual category The number of samples, The noisy label is The size of the sample set, For the noisy label In the sample of , the true label is The probability of A collection of categories.

8. A riverbed enhancement system integrating confidence learning and superpixel segmentation, characterized in that: The system implements the riverbed bottom quality enhancement method integrating confidence learning and superpixel segmentation as described in any one of claims 1 to 7, and the system comprises: A superpixel segmentation module, for segmenting the multi-beam backscatter intensity image using a superpixel segmentation method; A sample expansion module, used to expand the samples of pixels after superpixel segmentation; Confidence learning module, used to perform sample enhancement using confidence learning.

9. The riverbed quality enhancement system integrating confidence learning and superpixel segmentation according to claim 8 is characterized in that: The system is mounted on a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the processor executes the computer program, the functions of the riverbed sediment sample enhancement system integrating confidence learning and superpixel segmentation are realized.

10. The riverbed quality enhancement system integrating confidence learning and superpixel segmentation according to claim 8, characterized in that: The system is mounted on a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can realize the functions of a riverbed sediment sample enhancement system that integrates confidence learning and superpixel segmentation.

Images