Object-oriented seabed classification method and system based on backscatter intensity image

Abstract

The application provides an object-oriented seabed bottom type classification method and system based on a backscattering intensity image, sets weight parameters and scale parameters in combination with characteristics of the backscattering intensity image and local difference change rates; the FNEA algorithm is used to segment the backscattering intensity image; an object-oriented feature analysis method is used to process the segmented object, and multi-beam backscattering intensity image features are obtained; the random forest algorithm is used to classify the features with seabed sampling, and a seabed bottom type distribution map is obtained. The application gets rid of the traditional pixel-based feature extraction method, weakens the central anomaly of the multi-beam backscattering intensity image, eliminates the influence of pixel-level noise points, extracts features after segmentation of the intensity image, effectively maintains the contour integrity, and thus improves the accuracy of the seabed bottom type distribution map.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of geodetic surveying and mapping engineering, and particularly relates to an object-oriented seabed bottom classification method and system based on backscattering intensity images. BACKGROUND

[0002] Seafloor sediment distribution, as an important basic marine geographic information, not only serves the division of marine biological community boundaries, but also provides an important reference for the detection of ocean oil and gas, polymetallic nodules and other mineral resources. Therefore, studying the marine sediment classification method plays an important role in marine environmental protection and marine development.

[0003] The traditional method of obtaining bottom information by seabed sampling has problems such as low efficiency, high cost, and low resolution, resulting in low accuracy of bottom classification results. Since the 1960s, multi-beam sounding systems have become the main equipment for underwater topographic and geomorphic surveying after more than half a century of development. Multi-beam sounding systems can form dozens to hundreds of beams within a fan, achieving low-cost, high-efficiency, and full-coverage seafloor topographic surveying. The recorded seafloor backscattering intensity is related to the acoustic characteristics of the bottom, and can be applied to bottom classification research. Therefore, studying the multi-beam bottom classification method plays an important role in quickly obtaining high-precision bottom distribution information.

[0004] The light and dark degree of the multi-beam backscattering intensity image reflects the strength of the scattering ability of the seafloor bottom at the reference angle, and the image features are the main data products of multi-beam bottom classification. The complexity of the marine environment and other factors cause high-frequency noise and stripe noise in the intensity image, which seriously affects the effectiveness of the image features, and further affects the accuracy of the bottom classification. Therefore, the purpose of the present application is to propose an object-oriented feature analysis and bottom classification method to weaken the influence of backscattering intensity image noise on the accuracy of the classification results. SUMMARY

[0005] The present application aims to solve the problem of the complexity of the marine environment and other factors causing high-frequency noise and stripe noise in the intensity image, which seriously affects the effectiveness of the image features, and provides a backscattering intensity image feature accurate extraction and seabed bottom classification method combining FNEA algorithm and object-oriented analysis.

[0006] Based on the above technical problems, the present application adopts the following technical solutions:

[0007] An object-oriented seabed bottom classification method based on backscattering intensity images, comprising the following steps,

[0008] Step 1, obtaining a backscattering intensity image based on multi-beam backscattering intensity data;

[0009] Step 2, segment the backscatter intensity image obtained in step 1 using the FNEA algorithm;

[0010] Step 3, use an object-oriented feature extraction method to weaken the noise contribution in the object feature value, and perform feature extraction on the segmented backscatter intensity image in step 2 and obtain feature geometry;

[0011] Step 4, use a random forest algorithm to classify the feature set obtained in step 3 to realize seafloor bottom classification.

[0012] Further, in step 1, the specific process of obtaining the backscatter intensity image is: obtaining the backscatter intensity image by performing position calculation, radiation distortion correction, and angle response removal processing on the multi-wave backscatter intensity data.

[0013] Further, step 2 includes the following sub-steps:

[0014] Step 2.1, select the shape heterogeneity parameter, compactness parameter and scale parameter;

[0015] Step 2.2, set the shape heterogeneity parameter, compactness parameter and scale parameter in the FNEA algorithm, and segment the backscatter intensity image.

[0016] Further, the shape heterogeneity parameter and the compactness parameter in step 2.1 are selected in combination with the segmentation effect of different parameter values; the scale parameter is selected according to the local difference change rate curve, and the scale parameter corresponding to the peak value of the change rate curve is selected as the candidate scale. The segmentation effect of the candidate scale parameter is compared to select the best scale parameter.

[0017] Further, the local difference LV in step 2.1 is calculated as follows:

[0018]

[0019] Where N is the number of segmented objects, n i represents the number of pixels contained in the ith object, σ i represents the spectral standard deviation of the ith object;

[0020] The local difference change rate ROC-LV is calculated as follows:

[0021]

[0022] Where LV L is the local difference of the Lth object layer, and LV L-1 is the local difference of the object layer of the L-1 layer before the Lth layer.

[0023] Further, step 3 includes the following sub-steps:

[0024] Step 3.1, taking each object composed of pixel blocks in the segmented backscattering intensity image as a basic analysis unit of feature;

[0025] Step 3.2, extracting the mean and standard deviation of the gray scale of all pixels in each object;

[0026] Step 3.3, constructing a gray scale co-occurrence matrix for each object respectively;

[0027] Step 3.4, calculating the statistics of the gray scale co-occurrence matrix according to the gray scale co-occurrence matrix.

[0028] Further, the mean of gray scale in step 3.2 is calculated as follows:

[0029]

[0030] The standard deviation of gray scale is calculated as follows:

[0031]

[0032] Wherein, n is the number of pixel points contained in the object, g i represents the gray scale value of the i-th pixel point in the object.

[0033] Further, in step 3.3, when constructing the gray scale co-occurrence matrix, in order to reduce the boundary influence, the step length of the pixel pair is 1, and the pixels in the 8-neighborhood of the object boundary are considered when counting the pixel pairs.

[0034] Further, in step 3.4, the statistics of the gray scale co-occurrence matrix include: energy Ang, contrast Con, correlation Cor, heterogeneity Dis, entropy Ent and homogeneity Hom, and the calculation formulas are as follows:

[0035]

[0036] Con = ∑ i ∑ j (i-j) 2 P ij

[0037]

[0038] Dis = ∑ i ∑ j |i-j|·P ij

[0039] Ent = - ∑ i ∑ j P ij logP ij

[0040]

[0041] The calculation of μ1, μ2, σ1 and σ2 is as follows:

[0042] μ1 = ∑ i i∑ j P ij

[0043] μ2 = ∑ i j∑ j P ij

[0044] σ1 = ∑ i (i-μ1) 2 ∑ j P ij

[0045] σ2 = ∑ i (i-μ2) 2 ∑ j P ij

[0046] wherein i and j represent the gray scale of a pixel, and P ij is the probability of the pixel pair with pixel level i and j, respectively.

[0047] The application also provides an object-oriented seabed bottom classification system based on backscattering intensity images, comprising:

[0048] a backscattering image acquisition module for acquiring a backscattering intensity image based on multi-wave backscattering intensity data;

[0049] a segmentation module for performing segmentation processing on the backscattering intensity image acquired in step 1 by using an FNEA algorithm;

[0050] a feature extraction module for performing feature extraction on the backscattering intensity image segmented in the segmentation module and acquiring feature geometry by using an object-oriented feature extraction method to weaken the noise contribution in the object feature values;

[0051] a classification module for performing classification on the feature set obtained in the feature extraction module by using a random forest algorithm to realize seabed bottom classification.

[0052] Compared with the prior art, the application has the following beneficial effects:

[0053] The object-oriented seabed bottom type classification method based on the backscattering intensity image provided in the present application sets weight parameters and scale parameters in combination with the characteristics of the backscattering intensity image and the local difference change rate, uses the FNEA algorithm to segment the backscattering intensity image, uses the object-oriented feature analysis method to extract the object features of the backscattering intensity image, uses the random forest algorithm to classify the features with seabed sampling, and obtains a seabed bottom type distribution map. The seabed bottom type distribution map obtained by the present application has clear boundaries and few noise points, gets rid of the traditional pixel-oriented feature extraction method, weakens the influence of high-frequency noise and stripe noise, and improves the accuracy of the seabed bottom type distribution map. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A flowchart of an embodiment of the present application;

[0055] Figure 2 A comparison between the feature map obtained by the present application and the calculation result of the traditional pixel-based method;

[0056] Figure 3 A bottom type classification result of the traditional pixel-based method;

[0057] Figure 4 A bottom type classification result obtained by the present application. DETAILED DESCRIPTION

[0058] In order to make the technical problems to be solved by the embodiments of the present application, the technical solutions and beneficial effects more clearly understood, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0059] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the embodiments of the present application and simplify the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0060] In the description of the present application, unless otherwise specified, the term "connection" should be understood broadly, for example, it can be a fixed connection, a detachable connection, or an integral connection. For those skilled in the art, the specific meaning of the above-mentioned term in the present application can be understood according to the specific circumstances.

[0061] The implementation process of the present application will be further described in detail below in combination with specific drawings and examples.

[0062] Embodiment 1

[0063] The object-oriented seabed classification method based on the backscattering intensity image provided by the embodiment of the present application has the flow as shown in the figure Figure 1 .

[0064] Step 1: Based on the EM3002 measurement data of Jiaozhou Bay, the backscattering intensity image related only to the seabed characteristics is obtained after the standard multi-beam backscattering data processing step,

[0065] In the above embodiment, the multi-beam backscattering intensity data is processed to obtain the backscattering intensity image through the home position calculation, radiation distortion correction and angle response removal.

[0066] Step 2: The FNEA algorithm is used to segment the backscattering intensity image obtained in step 1.

[0067] Step 2.1: Select the shape heterogeneity parameter, compactness parameter and scale parameter.

[0068] Step 2.2: Set the shape heterogeneity parameter, compactness parameter and scale parameter in the FNEA algorithm to segment the backscattering intensity image.

[0069] The shape heterogeneity parameter and compactness parameter in step 2.1 are selected in combination with the segmentation effect of different parameter values; the scale parameter is selected according to the local difference change rate curve, and the scale parameter corresponding to the peak value of the change rate curve is selected as the alternative scale, and the segmentation effect of the alternative scale parameter is compared to select the best scale parameter.

[0070] The local difference LV in step 2.1 is calculated as follows:

[0071]

[0072] Wherein, N is the number of segmented objects, n i represents the number of pixels contained in the i-th object, σ i represents the spectral standard deviation of the i-th object.

[0073] The local difference change rate ROC-LV is calculated as follows:

[0074]

[0075] Wherein, LV L is the local difference of the L-th object layer, and LV L-1 is the local difference of the L-1-th object layer.

[0076] In the above embodiment, the shape heterogeneity parameter w shape is set to 0.6 and the compactness parameter w compThe peak value of the ROC-LV curve is selected as the candidate scale parameter, and the optimal scale parameter is 85 by combining the segmentation effect under different scale parameters.

[0077] In step 3, the object feature values are weakened by using an object-oriented feature extraction method, and the feature geometry is obtained by extracting features from the backscattering intensity image segmented in step 2.

[0078] The step 3 comprises the following sub-steps:

[0079] In step 3.1, each object composed of a pixel block in the segmented backscattering intensity image is taken as a basic unit for feature analysis.

[0080] In step 3.2, the average gray value and the standard deviation of all pixels in each object are extracted.

[0081] In step 3.3, a gray level co-occurrence matrix is constructed for each object.

[0082] In step 3.4, the statistical quantity of the gray level co-occurrence matrix is calculated according to the gray level co-occurrence matrix.

[0083] The average gray value in step 3.2 is calculated as follows:

[0084]

[0085] The standard deviation of gray value is calculated as follows:

[0086]

[0087] Wherein, n is the number of pixels contained in the object, g i represents the gray value of the i-th pixel in the object.

[0088] In step 3.3, when constructing the gray level co-occurrence matrix, in order to reduce the boundary effect, the pixel pair step is 1, and the pixels in the 8-neighborhood of the object boundary are considered when counting the pixel pairs.

[0089] The feature results of the present application and the conventional pixel-based method are compared as shown in the figure. Figure 2 As shown in the figure, subgraphs a, b and c are feature images obtained by using the conventional pixel-oriented analysis method, including pixel gray value, homogeneity and entropy; subgraphs d, e and f are feature images obtained by using the object-oriented analysis method of the present application, including object gray value, homogeneity and entropy. There are obvious noise points and linear noise in a, b and c, and there are no noise points and linear noise in d, e and f, so the present application can better suppress the influence of noise compared with the conventional method.

[0090] Step 4. The feature set obtained in step 3 is classified by using the random forest algorithm to realize the classification of the seafloor bottom.

[0091] The gray level co-occurrence matrix statistics calculated in step 3.4 include: energy Ang, contrast Con, correlation Cor, dissimilarity Dis, entropy Ent and homogeneity Hom, and the calculation formula is as follows:

[0092]

[0093] Con = ∑ i ∑ j (i-j) 2 P ij

[0094]

[0095] Dis = ∑ i ∑ j |i-j|·P ij

[0096] Ent = -∑ i ∑ j P ij logP ij

[0097]

[0098] The calculation of μ1, μ2, σ1, σ2 is as follows:

[0099] μ1 = ∑ i i∑ j P ij

[0100] μ2 = ∑ i j∑ j P ij

[0101] σ1 = ∑ i (i-μ1) 2 ∑ j P ij

[0102] σ2 = ∑ i (i-μ2) 2 ∑ j P ij

[0103] Wherein, i, j represent the gray level of the pixel point, P ij is the probability of the pixel pair with pixel level i and j respectively.

[0104] The traditional bottom classification result based on pixel method is as follows Figure 3As shown, there are a large number of noise points and serious misclassification caused by the central anomaly of the strip in the bottom type distribution map. The bottom type classification result obtained by the present application is as follows Figure 4 There are no obvious noise points in the bottom type distribution map, and there is no misclassification caused by the central anomaly of the strip, proving that the bottom type classification result of the present application has strong reliability and strong noise suppression capability.

[0105] Embodiment 2

[0106] The present embodiment provides an object-oriented seabed bottom type classification system based on backscattering intensity images, comprising:

[0107] A backscattering image acquisition module acquires a backscattering intensity image based on multi-wave backscattering intensity data;

[0108] A segmentation module performs segmentation processing on the backscattering intensity image acquired in step 1 using an FNEA algorithm;

[0109] A feature extraction module uses an object-oriented feature extraction method to weaken the noise contribution in the object feature values, performs feature extraction on the segmented backscattering intensity image in the segmentation module, and acquires feature geometry;

[0110] A classification module classifies the feature set obtained in the feature extraction module using a random forest algorithm to realize seabed bottom type classification.

[0111] The above is only the preferred embodiment of the present application, and does not limit the implementation and protection scope of the present application. For those skilled in the art, it should be realized that any equivalent replacement and obvious changes made according to the content of the present application should be included in the protection scope of the present application.

Claims

1. An object-oriented seabed classification method based on backscatter intensity images, characterized in that: The method comprises the following steps, Step 1, obtaining a backscattering intensity image based on multi-wave backscattering intensity data; Step 2, performing segmentation processing on the backscattering intensity image obtained in the step 1 by using an FNEA algorithm; Step 3, performing feature extraction on the backscattering intensity image segmented in the step 2 by using an object-oriented feature extraction method and obtaining feature geometry, including the following sub-steps: Step 3.1, taking each object composed of a pixel block in the segmented backscattering intensity image as a basic unit for feature analysis; Step 3.2, extracting the mean and standard deviation of the gray scale of all pixels in each object; the mean of the gray scale is calculated as follows: The standard deviation of the gray scale is calculated as follows: wherein, n is the number of pixel points contained by the object, g i denotes the gray value of the pixel point in the object, i the object. Step 3.3, constructing a gray level co-occurrence matrix for each object respectively; when constructing the gray level co-occurrence matrix, in order to reduce the boundary effect, the step of the pixel pair is selected to be 1, and the pixels in the 8-neighborhood of the object boundary are considered when counting the pixel pairs; Step 3.4, according to the gray level co-occurrence matrix, the statistical quantity of the calculated gray level co-occurrence matrix respectively; the calculated gray level co-occurrence matrix statistics include: energy Ang , contrast Con , correlation Cor , heterogeneity Dis , entropy Ent and homogeneity Hom , the calculation formula is as follows: μ 1、 μ 2、 σ 1、 σ 2 is calculated as follows: in, i , j Represents the gray level of a pixel. P ij The pixel level of the pixel pair is respectively i and j The probability of; Step 4, classifying the feature set obtained in the step 3 by using a random forest algorithm to realize seabed bottom classification.

2. The object-oriented seabed classification method based on backscatter intensity image according to claim 1, characterized in that: In the step 1, the specific process of obtaining the backscattering intensity image is as follows: the multi-wave backscattering intensity data is subjected to homing calculation, radiation distortion correction and angle response removal processing to obtain the backscattering intensity image.

3. The object-oriented seabed classification method based on backscatter intensity image according to claim 1, characterized in that: The step 2 comprises the following sub-steps: Step 2.1, selecting a shape heterogeneity parameter, a compactness parameter and a scale parameter; Step 2.2, setting the shape heterogeneity parameter, the compactness parameter and the scale parameter in the FNEA algorithm to segment the backscattering intensity image.

4. The object-oriented seabed classification method based on backscatter intensity image according to claim 3, characterized in that: The selection of the shape heterogeneity parameter and the compactness parameter in the step 2.1 is to be combined with the segmentation effect of different parameter values; the scale parameter is to be selected according to the change rate curve of the local difference, and the scale parameter corresponding to the peak value of the change rate curve is taken as the candidate scale, and the best scale parameter is selected by comparing the segmentation effects of the candidate scale parameters.

5. The object-based seabed classification method based on backscatter intensity image according to claim 4, characterized in that: The local difference LV in the step 2.1 is calculated as follows: wherein, N is the number of segmented objects, n i represents the number of pixels contained in the i th object, The change rate ROC-LV of the local difference is calculated as follows: i represents the spectral standard deviation of the i th object; It comprises: Among them, LV L It is the first L Local differences in the object layer, LV L-1 It is the first L The layer before the layer, that is L Local differences in the object layer at level -1.

6. An object-oriented seafloor bottom type classification system based on backscatter intensity images, characterized by, a backscattering image acquisition module which obtains a backscattering intensity image based on multi-wave backscattering intensity data; a segmentation module which performs segmentation processing on the backscattering intensity image obtained in the step 1 by using an FNEA algorithm; a feature extraction module which performs feature extraction on the backscattering intensity image segmented in the segmentation module by using an object-oriented feature extraction method and obtains feature geometry; a classification module which classifies the feature set obtained in the feature extraction module by using a random forest algorithm to realize seabed bottom classification; The object-oriented seabed bottom classification system based on the backscattering intensity image is used to execute the steps in the object-oriented seabed bottom classification method based on the backscattering intensity image in any one of claims 1-5. ​

Images