Automatic classification and mapping method of seabed sediment type

Abstract

The present application belongs to the field of classification of seabed sediment types, and specifically relates to a method for automatically classifying and mapping seabed sediment types, comprising the following steps: collecting acoustic image data of seabed sediments, pre-processing and sound intensity compensation correction of the acoustic image data; analyzing the pre-processed acoustic image data, extracting image features, and constructing a seabed sediment classification model for automatic classification; the present application can efficiently realize the investigation and research of large-scale seabed sediment types by automatically classifying seabed sediment types and generating distribution maps, can significantly improve the accuracy and efficiency of sediment type classification, and can further improve the accuracy of classification, greatly speed up the classification speed, and reduce the time and labor cost by using advanced image processing and machine learning technology to automatically extract key features in acoustic images and learn the mapping relationship between sediment types and features based on a large amount of training data.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of classification of seabed sediment types, and particularly relates to a method for automatically classifying and mapping seabed sediment types. BACKGROUND

[0002] Seabed sediments are an important part of the marine environment, and their types, distribution and changes are of great significance to understanding marine ecosystems, seabed topography and marine geological processes. Seabed sediments contain rich mineral resources such as oil, natural gas and mineral sands. Accurate seabed sediment type information is crucial for resource exploration and development. Through automatic classification and mapping, the type information of seabed sediments can be efficiently and accurately obtained, thereby providing important data support for marine scientific research.

[0003] In the prior art, different types of seabed sediments may have class imbalance, and the small amount of data samples may result in poor classification effect on a small number of classes, thereby affecting the accuracy of the overall classification result. Therefore, how to improve the quality of seabed sediment exploration data and improve the accuracy of the classification and mapping result is a problem to be solved at present. Therefore, a method for automatically classifying and mapping seabed sediment types is proposed. SUMMARY

[0004] To solve the problem of poor classification effect of existing sediments, the application provides a method for automatically classifying and mapping seabed sediment types. By using image processing and machine learning technology, key feature information in acoustic images is automatically extracted, a mapping relationship between sediment types and features is determined based on a classification model, and seabed sediment types are automatically classified and distribution maps are generated.

[0005] The application is implemented by the following technical scheme, a method for automatically classifying and mapping seabed sediment types, comprising the following steps:

[0006] Step A, collect acoustic image data of seabed sediments, and perform data processing on the acoustic image data to improve data quality;

[0007] Step B, analyze the processed acoustic image data and extract image features;

[0008] Step C, construct a seabed sediment classification model and train it, input the processed acoustic image data and extracted image features into the classification model, and analyze the correlation between each feature and the seabed sediment type;

[0009] Step D, based on the trained sediment classification model, input the acoustic image data to be classified, predict the seabed sediment type, and convert the classification result into an image format to generate a seabed sediment type distribution map.

[0010] Further, in the step A, the data processing operation includes pre-processing operations such as denoising, geometric correction, and gray scale stretching of the acoustic image data. The denoising process removes noise in the acoustic image through a digital filter, and then the acoustic image is geometrically corrected to correct the scale and rotation angle of the image, ensure the correct geometric shape and proportion of the image, and finally the image contrast and brightness are adjusted through gray scale stretching processing to make the features in the image more clear and visible.

[0011] Further, in the step A, the acoustic intensity compensation correction is performed on the pre-processed acoustic image data, the attenuation law of acoustic intensity with increasing sound propagation distance is analyzed, and the acoustic intensity is compensated according to the sound propagation distance to reflect the real physical characteristics of the seabed sediments and eliminate the acoustic intensity changes caused by different sound propagation distances. The signal-to-noise ratio and contrast of the processed acoustic image data are evaluated to check whether the image quality meets the standard.

[0012] The acoustic intensity compensation correction process is implemented in the following manner:

[0013] (1) According to the acoustic attenuation law, a mathematical model between acoustic intensity and distance is established to determine the acoustic intensity compensation correction model and obtain the acoustic intensity compensation correction coefficient, which is represented as:

[0014] I = I0·e -Ad ;

[0015]

[0016] where I is the acoustic intensity compensation correction coefficient, I represents the acoustic intensity at a distance d from the sound source, I0 is the acoustic intensity at the sound source, A is the attenuation coefficient of sound, d is the sound propagation distance, and d0 is the reference distance.

[0017] (2) The pre-processed acoustic image data is read, and the sound propagation distance corresponding to each pixel point is calculated based on the parameters of the sonar equipment and the position information of the pixel points in the image. The sound propagation distance corresponding to each pixel point is substituted into the acoustic intensity compensation correction coefficient to calculate the compensated acoustic intensity value.

[0018] (3) The compensated acoustic intensity value is substituted into the original acoustic image data to obtain the acoustic image data after acoustic intensity compensation correction.

[0019] Further, in the step A, the signal-to-noise ratio and contrast of the processed acoustic image data are evaluated in the following manner:

[0020] (1) Based on the pre-processed acoustic image data, the average power data of the signal and the average power data of the noise in the image are extracted to obtain the signal-to-noise ratio evaluation index, which is represented as:

[0021]

[0022] Wherein, SR is the signal-to-noise ratio evaluation index, used to measure the ratio of signal to noise in the image to evaluate the image quality, SP is the average power of the signal in the image, the signal refers to the characteristic area of the seabed sediment, NP is the average power of the noise;

[0023] (2) According to the preprocessed acoustic image data, the gray level number and the gray mean data of the image are determined, and the contrast evaluation index is obtained, and the contrast evaluation index is represented as:

[0024]

[0025] Wherein, C is the contrast evaluation index, reflecting the brightness difference between different regions in the image, L is the gray level number of the image, usually 256, μ is the gray mean, P(i) represents the probability of the pixel with the gray value i

[0026] (3) The signal-to-noise ratio evaluation index and the contrast evaluation index are integrated to evaluate the quality of the signal-to-noise ratio and the contrast of the processed acoustic image data.

[0027] Further, in step B, the extraction process of the image features includes:

[0028] (1) The Canny edge detection algorithm is used to detect the edge of the acoustic image data, extract the seabed sediment edge information in the preprocessed acoustic image data, and quantize the detected edge information, convert the edge information into numerical features, and extract the length, direction and curvature features of the edge information as edge features;

[0029] (2) The texture analysis method of the gray level co-occurrence matrix is used to analyze the texture of the preprocessed acoustic image data, extract the texture information of the seabed sediment, and take the mean, variance, energy, contrast and correlation of the texture information as the texture features;

[0030] (3) The features extracted from the edge information and the texture information are combined into a feature vector, each feature vector contains the edge and texture feature information in the preprocessed acoustic image data, which is used for training and classification of the machine learning model.

[0031] Further, in step C, the process of constructing the seabed sediment classification model and analyzing the correlation between each feature and the seabed sediment type is as follows:

[0032] (1) The historical acoustic image data processed in step A and the combined features obtained in step B are integrated into a feature vector as the input of the model, and the data set is divided into a training set and a test set;

[0033] (2) input the prepared data into the selected neural network model for training, and test the trained classification model using the test set to obtain the trained seabed sediment classification model;

[0034] (3) combine the pre-processed historical acoustic image data and the seabed sediment classification model to obtain a correlation analysis coefficient, observe the correlation degree of each feature in the classification with the seabed sediment type, and determine the importance of each feature.

[0035] wherein the correlation analysis coefficient is:

[0036]

[0037] wherein R 2 is the correlation analysis coefficient, r is the number of categories of the row variable, indicating the number of different types of seabed sediments, c is the number of categories of the column variable, indicating the number of different values of a certain feature, O jk is the observation frequency, indicating the number of times that the jth category of the row variable and the kth category of the column variable appear simultaneously in the actual data, and E jk is the expected frequency, indicating the number of times that the jth category of the row variable and the kth category of the column variable appear simultaneously if the row variable and the column variable are independent, according to the total number of rows and columns.

[0038] Further, in the step D, when the seabed sediment type distribution map is generated, the classification prediction results output by the seabed sediment classification model are arranged to form a data set containing the seabed sediment type of each position; the data set is converted into an image format using GIS to generate the seabed sediment type distribution map, and different colors are used to represent different sediment types in the seabed sediment type distribution map, and the sediment types of different regions are labeled to visually display the spatial distribution of seabed sediments.

[0039] Compared with the prior art, the advantages and positive effects of the present application are:

[0040] The present application utilizes advanced image processing and machine learning techniques to automatically extract key features from acoustic images, learns the mapping relationship between sediment types and features based on a large amount of data, and significantly improves the accuracy and efficiency of sediment type classification by automatically classifying seabed sediment types and generating distribution maps, and efficiently realizes the investigation and research of large-scale seabed sediment types; by regularly collecting acoustic image data and applying the automatic classification method, the changes of seabed sediments can be continuously monitored, and the related change trend of sediment types can be understood;

[0041] In addition, the classification results are intuitively displayed in the form of images through automatic classification into a map, so that the seabed sediment types are intuitively understood and analyzed, the distribution of different sediment types on the seabed, and the spatial relationship and variation law between the different sediment types can be clearly determined through the generated sediment type distribution map, the marine environment is better understood, potential resources and ecological environment problems are found, and the data analysis and decision-making efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The method flowchart of the embodiment of the present application is described.

[0043] Figure 2 The collection and processing flowchart of the acoustic image data of the seabed sediment of the embodiment of the present application is described.

[0044] Figure 3 The sound intensity compensation correction flowchart of the preprocessed acoustic image data of the embodiment of the present application is described.

[0045] Figure 4 The construction of the seabed sediment classification model and the correlation analysis of each feature and the seabed sediment type flowchart of the embodiment of the present application is described.

[0046] Figure 5 The generation flowchart of the seabed sediment type distribution map of the embodiment of the present application is described. DETAILED DESCRIPTION

[0047] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is further described below in combination with the drawings and embodiments. In the following description, a large number of specific details are set forth in order to facilitate a full understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the present application is not limited to the specific embodiments disclosed below

[0048] The embodiment, as shown in the drawings, proposes an automatic classification into a map method of seabed sediment types, which comprises the following steps: Figure 1

[0049] Step A, collecting acoustic image data of seabed sediments, and performing data processing on the acoustic image data to improve data quality;

[0050] Step B, analyzing the processed acoustic image data, and extracting image features;

[0051] Step C, constructing a seabed sediment classification model and training the same, inputting the processed acoustic image data and the extracted image features into the classification model, and analyzing the correlation between each feature and the seabed sediment type;

[0052] ​Step D, based on the trained sediment classification model, input the acoustic image data to be classified, classify and predict the seabed sediment type, and convert the classification result into an image format to generate a seabed sediment type distribution map.

[0053] Specifically, in combination with Figures 2-3 As shown in the figure, in step A, acoustic image data of seabed sediments is acquired, and the acoustic image data is preprocessed and sound intensity compensation is corrected to improve data quality.

[0054] Step A1, collect acoustic image data and pre-process it;

[0055] The acoustic sensors of the sonar equipment are used to scan the seabed and acquire acoustic image data, wherein the acoustic image data includes historical acoustic image data and real-time acoustic image data, and the collected data covers different types of seabed sediments to facilitate subsequent analysis; wherein the historical acoustic image data refers to the acoustic image data acquired in the past time period. Here, the historical acoustic image data and the real-time acoustic image data are distinguished in order to distinguish the acquisition period of the data.

[0056] The acoustic image data is pre-processed, such as denoising, geometric correction, and gray scale stretching, to improve data quality; wherein the denoising process removes noise in the acoustic image through a digital filter, then geometric correction is performed to correct the scale and rotation angle of the image, ensuring the geometric shape and proportion of the image, and finally the image is processed through gray scale stretching to adjust the contrast and brightness of the image, making the features in the image clearer.

[0057] Step A2, sound intensity compensation correction is performed on the pre-processed data;

[0058] (1) The pre-processed acoustic image data is subjected to sound intensity compensation correction, the attenuation law of sound intensity with increasing sound propagation distance is analyzed, and the sound intensity is compensated according to the distance of sound propagation to reflect the true physical characteristics of seabed sediments and eliminate the sound intensity changes caused by different sound propagation distances.

[0059] The sound intensity compensation correction process includes:

[0060] According to the sound attenuation law, a mathematical model between sound intensity and distance is established to determine the sound intensity compensation correction model and obtain the sound intensity compensation correction coefficient; the sound intensity compensation correction coefficient is expressed as follows:

[0061] I = I0·e -Ad ;

[0062]

[0063] Wherein, I is the sound intensity compensation correction coefficient, indicates the sound intensity at the distance of the sound source d, I0 is the sound intensity at the sound source, A is the attenuation coefficient of sound, d is the sound propagation distance, d0 is the reference distance;

[0064] Then the pre-processed acoustic image data is read, the sound wave propagation distance corresponding to each pixel point is calculated according to the parameters of the sonar equipment and the position information of the pixel points in the image, the sound wave propagation distance corresponding to each pixel point is substituted into the sound intensity compensation correction coefficient, and the compensated sound intensity value is calculated. Replace the sound intensity value of the corresponding pixel point in the original acoustic image data with the calculated compensated sound intensity value to obtain the acoustic image data after sound intensity compensation correction.

[0065] (2) Perform signal-to-noise ratio and contrast evaluation to check whether the image quality meets the standard;

[0066] For the data after sound intensity compensation correction, the average power data of the signal and the average power data of the noise in the image are extracted to obtain the signal-to-noise ratio evaluation index; and the gray level number and the gray mean data of the image are determined to obtain the contrast evaluation index; the signal-to-noise ratio evaluation index and the contrast evaluation index are combined to evaluate the image data quality;

[0067] Wherein, the signal-to-noise ratio evaluation index is:

[0068]

[0069] Wherein, SR is the signal-to-noise ratio evaluation index, which is used to measure the ratio of signal to noise in the image to evaluate the image quality, SP is the average power of the signal in the image, the signal refers to the characteristic area of the seabed sediment, NP is the average power of the noise, and it should be noted that the higher the value of the signal-to-noise ratio evaluation index, the better the image quality, and the value of SR needs to be greater than 20dB;

[0070] The contrast evaluation index is:

[0071]

[0072] Wherein, C is the contrast evaluation index, which reflects the brightness difference between different regions in the image, L is the gray level number of the image, which is usually 256, μ is the gray mean, and P(i) represents the probability of the occurrence of the pixel with gray value i. It should be noted that the higher the contrast evaluation index, the greater the brightness difference between different regions in the image, and the clearer the image.

[0073] In step B, for the data after sound intensity compensation correction, the specific process of analyzing and extracting the edge features and texture features of the image is as follows:

[0074] Step B1, based on the Canny edge detection algorithm, the edge of the acoustic image data is detected, the edge information of the seabed sediment in the preprocessed acoustic image data is extracted, and the detected edge information is quantized, the edge information is converted into numerical features, and the length, direction and bending degree features of the edge information are extracted as edge features;

[0075] Step B2, based on the texture analysis method of gray level co-occurrence matrix, the texture of the seabed sediment is analyzed, and the texture information is extracted, and the mean, variance, energy, contrast and correlation of the texture information are used as texture features;

[0076] Step B3, the edge features and texture features extracted from the edge information and texture information are combined into feature vectors, which are used for training and classification of machine learning models;

[0077] In combination Figure 4 As shown in the embodiment, in step C, the seabed sediment classification model is constructed, the acoustic image data processed in step A and the combined feature vectors selected in step B are input into the classification model, the correlation of each feature with the seabed sediment type is analyzed, and the specific process is as follows:

[0078] The feature vectors combined by the edge features and the texture features are used as the input of the classification model, the data set is divided into training set and test set, the seabed sediment classification model is trained, and the trained classification model is tested and evaluated using the test set, and the trained seabed sediment classification model is obtained;

[0079] In combination with historical acoustic image data and seabed sediment classification model, the correlation analysis coefficient is obtained, the correlation degree of each feature in classification with the seabed sediment type is observed, and the importance of each feature is determined;

[0080] The calculation formula of the correlation analysis coefficient is as follows:

[0081]

[0082] Wherein, R 2 is the correlation analysis coefficient, r is the number of categories of the row variable, indicating the number of different types of seabed sediments, c is the number of categories of the column variable, indicating the number of different values of a feature, O jk is the observation frequency, indicating the number of times that the jth category of the row variable and the kth category of the column variable appear at the same time in the actual data, E jk is the expected frequency, indicating the number of times that the jth category of the row variable and the kth category of the column variable appear at the same time if the row variable and the column variable are independent, according to the total number of rows and columns.

[0083] It should be noted that when R 2When the value of R is large, it means that the difference between the observed data and the expected data is large, that is, there is a strong correlation between the feature and the seafloor sediment type; when the value of R 2 is small, it means that the difference between the observed data and the expected data is not large, which means that the correlation between the feature and the seafloor sediment type is weak.

[0084] Finally, the seafloor sediment classification model is deployed, the acoustic image data to be classified is input, the seafloor sediment type is classified and predicted, and the classification result is converted into an image format to generate a seafloor sediment type distribution map. The specific process is as shown in Figure 5

[0085] The trained seafloor sediment classification model is deployed in a cloud platform, real-time acoustic image data after preprocessing is input into the seafloor sediment classification model for prediction, and the corresponding seafloor sediment type prediction result is output. The classification prediction result output by the seafloor sediment classification model is arranged to form a data set containing the seafloor sediment type of each position. The data set is converted into an image format using GIS to generate a seafloor sediment type distribution map. In the seafloor sediment type distribution map, different colors are used to represent different sediment types, and the sediment types of different regions are labeled to intuitively show the spatial distribution of the seafloor sediment.

[0086] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in other forms. Any skilled person in the art can modify or change the above disclosed technical content to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments shall still fall within the protection scope of the present application.​

Claims

1. A method for automatic classification and mapping of seabed sediment types, characterized in that, Includes the following steps: Step A: Acquire acoustic image data of seabed sediments, and perform preprocessing and sound intensity compensation correction on the acoustic image data; Step B: Analyze and extract the edge and texture features of the image based on the data after sound intensity compensation correction; Step C: Construct a seabed sediment classification model and train it. Input the acoustic image data after sound intensity compensation correction in Step A and the edge features and texture features obtained in Step B into the classification model, and analyze the correlation between each feature and the seabed sediment type. Step D: Input the acoustic image data to be classified, predict the type of seabed sediment based on the trained seabed sediment classification model, convert the classification results into image format, and generate a seabed sediment type distribution map. Step A specifically includes the following steps: Step A1: Acquire acoustic image data and perform denoising, geometric correction, and grayscale stretching preprocessing operations on it; the acoustic image data includes historical acoustic image data and real-time acoustic image data; Step A2: Perform sound intensity compensation correction on the preprocessed acoustic image data; (1) Analyze the attenuation law of sound intensity as the propagation distance of sound waves increases, and compensate for the sound intensity according to the propagation distance of sound waves to reflect the true physical characteristics of seabed sediments; Based on the sound wave attenuation law, a mathematical model is established between sound intensity and distance, the sound intensity compensation correction model is determined, and the sound intensity compensation correction coefficients are obtained. The sound intensity compensation correction coefficients are expressed as follows: ； ； in, The sound intensity compensation correction factor represents the distance from the sound source. The sound intensity at that location, The sound intensity at the sound source. The attenuation factor of the sound. For the distance the sound wave travels, For reference distance; Then, based on the processed acoustic image data, according to the parameters of the sonar device and the position information of the pixels in the image, the sound wave propagation distance corresponding to each pixel is calculated. The sound wave propagation distance corresponding to each pixel is substituted into the sound intensity compensation correction coefficient to calculate the compensated sound intensity value. The calculated compensated sound intensity value replaces the sound intensity value of the corresponding pixel in the original acoustic image data to obtain the acoustic image data after sound intensity compensation correction. (2) Then perform signal-to-noise ratio and contrast evaluation to check whether the image quality meets the standards; In step C, historical acoustic image data and seabed sediment classification models are combined to obtain correlation analysis coefficients, observe the degree of correlation between each feature and seabed sediment type in classification, and determine the importance of each feature. The formula for calculating the correlation coefficient is as follows: ； in, The correlation coefficient is the coefficient used in the correlation analysis. The number of categories for the row variable represents the number of different types of seabed sediments. The number of categories for a column variable represents the number of distinct values ​​for a particular feature. 'Observation frequency' indicates the number of rows in the actual data where the variable is observed. The first category and column variable The number of times each category appears simultaneously Let be the expected frequency, representing the predicted frequency of the row variable based on the total number of rows and columns if the row and column variables are independent. The first category and column variable The number of times each category appears simultaneously.

2. The method for automatic classification and mapping of seabed sediment types according to claim 1, characterized in that: In step A, the signal-to-noise ratio and contrast evaluation are performed using the following methods: For the data after sound intensity compensation correction, the average power data of the signal and the average power data of the noise in the image are extracted to obtain the signal-to-noise ratio evaluation index; the gray level and gray mean data of the image are determined and combined to obtain the contrast evaluation index; the image data quality is evaluated by combining the signal-to-noise ratio evaluation index and the contrast evaluation index. The signal-to-noise ratio evaluation index is as follows: ； in, This is a signal-to-noise ratio (SNR) evaluation index. This represents the average power of the signal in the image, where the signal refers to a characteristic region of the seabed sediment. The average power of the noise; The contrast ratio evaluation index is: ； in, This is the contrast ratio evaluation index. The gray level of the image. The grayscale mean is... Indicates grayscale value The probability of a pixel appearing.

3. The method for automatic classification and mapping of seabed sediment types according to claim 1, characterized in that: Step B is implemented in the following manner: Step B1: Perform edge detection on acoustic image data based on the Canny edge detection algorithm, extract edge information of seabed sediments in the preprocessed acoustic image data, and perform feature quantization on the detected edge information to convert the edge information into numerical features. Extract the length, direction, and curvature features of the edge information as edge features. Step B2: The texture analysis method based on gray-level co-occurrence matrix is ​​used to perform texture analysis on the sound intensity compensation and correction data, extract the texture information of seabed sediments, and use the mean, variance, energy, contrast and correlation of the texture information as texture features. Step B3: Combine the edge features and texture features extracted from the edge information and texture information into a feature vector, which is used for training and classification of the machine learning model.

4. The method for automatic classification and mapping of seabed sediment types according to claim 1, characterized in that: In step C, the feature vector composed of edge features and texture features is used as the input of the classification model. The dataset is divided into a training set and a test set. The seabed sediment classification model is trained and tested and evaluated using the test set to obtain the trained seabed sediment classification model.

5. The method for automatic classification and mapping of seabed sediment types according to claim 1, characterized in that: In step D, the process of generating the seabed sediment type distribution map is as follows: Based on the trained sediment classification model, the sediment micro-classification prediction results are output, forming a dataset containing the seabed sediment types for each location. The dataset is then converted into an image format using GIS to generate a seabed sediment type distribution map. In the seabed sediment type distribution map, different sediment types are represented by different colors, and the sediment types in different areas are labeled.

Images