non-Gaussian submarine landform type identification method based on multi-fractal spectral features
Technical Field
The invention relates to the technical fields of ocean mapping, ocean engineering, ocean oil and gas resources and the like, in particular to a non-Gaussian submarine landform type identification method based on multi-fractal spectrum characteristics.
Background
Topography is the general term for various relief forms on the earth's surface. Types of submarine topography include ancient watercourses, scour troughs, submarine canyons, deep water channels, sea mountains, carbonate terraces, scarps, landslides, and the like. The landform classification is the basis of submarine landform research and drawing, and the landform form can reflect the internal relation between the landform cause type and the cause control form, and is the key for deep-layer geological knowledge mining.
The multi-beam sounding system can detect the submarine topography and landform fine characteristics with wide coverage and high resolution. The system adopts strip type measurement, can simultaneously measure the water depth value of dozens or even hundreds of seabed measured points in a plane vertical to the direction of the flight line or a full-coverage water depth strip with a certain width, and can accurately and quickly measure the size, the shape and the height change of an underwater target in a certain width along the flight line.
The landform parameters are digital descriptions of landforms and are used for characterizing the spatial distribution of the landforms. Geomorphic parameters are numerous and vary in understanding and categorizing them in different disciplines and areas. In the current research, the geomorphic parameters mainly relate to two major categories, namely microscopic parameters and macroscopic parameters. The microscopic parameters describe and reflect topographical features of a particular location. The commonly used microscopic parameters mainly include gradient, slope direction, slope length, plane curvature, section curvature, etc. The macroscopic parameters describe and reflect topographical features over a large area. Common macro parameters mainly comprise terrain depth standard deviation, terrain difference entropy, terrain roughness, elevation variation coefficient and the like.
At the end of the 20 th century, Mandelbrot, a French mathematician, proposed a concept of fractal to represent complex processes or graphs. Fractal is a general term for self-similar patterns and structures with certain meaning but without characteristic length. Fractal structures are widely found in nature, and for example, coastlines, snowflakes, trees, clouds, and the like all have fractal structures. Multi-fractal is a measure of a two-or three-dimensional object that has spatial self-similarity or statistical self-similarity. The multi-fractal structure has non-Gaussian property, and the multi-fractal spectrum can reflect the complexity of the object structure.
The submarine landforms are complex, at present, the recognition of submarine landform types is limited, a large amount of landform type analysis is mainly completed manually, and the submarine landform types are distinguished and identified through observation and experience of technicians. The method can fully utilize knowledge of technicians, has good flexibility, but needs the technicians to have rich geoscience knowledge and observation and judgment experience, has great subjectivity, and has the defects of poor timeliness, high labor intensity and the like. Particularly, because the submarine data has a mass level, the processing task of the mass data cannot be born by the manual capability of technicians.
Disclosure of Invention
The method determines whether the terrain depth distribution has non-Gaussian property according to the kurtosis and skewness of the submarine terrain, extracts the space structure characteristics of fractal spectrum features representing the terrain of the non-Gaussian terrain, extracts the landform factors reflecting the landform types by applying a factor analysis method, and establishes a landform type classifier through a support vector machine to realize landform type identification. The extracted multi-fractal spectrum characteristics comprise multi-fractal spectrum width, multi-fractal spectrum peak value, maximum and minimum fractal dimension difference, capacity dimension and correlation dimension. The method has the advantages of simplicity, small calculated amount, high identification accuracy, manpower saving and the like. The method is suitable for identifying the type of the non-Gaussian submarine landform.
The invention comprises the following steps:
(1) judging non-Gaussian terrain:
are respectively based on
Calculating the skewness m of the seabed depth of the area to be identified
1Kurtosis m
2If m is satisfied
1> 0 and m
2If the terrain is more than 0, the terrain of the area to be identified is non-Gaussian type terrain, wherein z
iFor the normalized depth value of the sea floor at the ith measuring point in the area to be identified,
the average value of the depth of the sea bottom is obtained, and n is the number of measurement points in the area to be identified;
(2) calculating the characteristics of the multi-fractal spectrum:
according to respective Δ α ═ αmax-αmin、f(α(q))max=max(f(α(q)))、Δf=f(αmax)-f(αmin) And DqCalculating fractal spectrum width Δ α, fractal spectrum peak f (α (q))maxMaximum and minimum fractal dimension difference delta f and dimension DqWherein, when q is 0, DqFor capacity dimension, q is 2 hours DqFor the correlation dimension, max (f (α (q))) represents taking the maximum value of f (α (q)), α (q) being the Hausdor dimension of the topographic depth distribution of the area to be identified, αmaxIs the maximum singularity index, alphaminF (alpha (q)) is a distribution density function, alphamax、αminAlpha (q) and f (alpha (q)) are obtained by calculation by using a Chhabra algorithm, wherein q is a scale parameter and is set in the Chhabra algorithm;
(3) extracting landform factors based on multi-fractal spectrum characteristics:
original variables are Δ α, f (α (q))
maxΔ f and D
qExtracting a common factor f of the original variables by using a factor analysis method
iI is 1,2,3, … …, the variance contribution ratio of each factor is λ
iI is 1,2,3, … …, if satisfied
F is then
iI is 1,2, … …, and is a landform factor based on the multi-fractal spectrum characteristics, wherein t is the number of common factors, f
i、λ
iT is determined according to factor analysis, theta is a threshold value and is set in program parameters, and j is the number of topographic factors;
(4) and (3) designing a landform type classifier:
according to the landform factor f i1,2, … …, applying a support vector machine to determine a landform type classifier;
(5) and (3) landform type identification:
calculating the skewness and kurtosis value m of the terrain depth of the landform to be identified1And m2If m is satisfied1> 0 and m2If the number is more than 0, calculating the multi-fractal spectrum characteristics of the landform to be identified and the landform factor value f thereofiAnd (5) determining the type of the landform to be identified according to the classifier obtained in the step (4), wherein i is 1,2 and … ….
Drawings
Fig. 1(a) to 1(e) are topographical views of 5 types of terraces, including a seabed terrace, a gully, a landslide, a ridge, and a water course, respectively;
FIG. 2 is a cross-sectional view of skewness and kurtosis of the depth distribution of 5 kinds of landforms;
FIGS. 3(a) to 3(f) are cross-sectional views of fractal spectral features of non-Gaussian landforms in this embodiment;
FIG. 4 is a plot of the variance contribution of 5 common factors for 5 multifractal spectral features of a non-Gaussian landform type;
FIG. 5 is a plot of the cumulative variance contribution of 5 common factors for 5 multifractal spectral features of a non-Gaussian landform type;
FIG. 6 is a non-Gaussian landform type classifier in accordance with the present embodiment;
fig. 7 shows the recognition result of the non-gaussian landform type in this embodiment.
Detailed Description
According to the embodiment, according to multi-beam seabed sounding data, the kurtosis and skewness values of seabed terrain are calculated to determine whether the terrain depth distribution has non-Gaussian nature, the multi-fractal spectrum features and the landform factors of the non-Gaussian terrain are extracted, and a landform type classifier is established through a support vector machine to realize the identification of 3 non-Gaussian landform types such as seabed gully, landslide and water channel. The extracted multi-fractal spectrum characteristics comprise multi-fractal spectrum width, multi-fractal spectrum peak value, maximum and minimum fractal dimension difference, capacity dimension and correlation dimension.
The specific identification steps are as follows:
(1) judging non-Gaussian terrain:
are respectively based on
Calculating the skewness m of the seabed depth of the area to be identified
1Kurtosis m
2If m is satisfied
1> 0 and m
2If the terrain is more than 0, the terrain of the area to be identified is non-Gaussian type terrain, wherein z
iFor the normalized depth value of the sea floor at the ith measuring point in the area to be identified,
the average value of the depth of the sea bottom is obtained, and n is the number of measurement points in the area to be identified.
In the present embodiment, the seafloor depth data is measured by a multi-beam method, and fig. 1(a) to 1(e) are topographic maps of 5 types of seafloor plateaus, gullies, landslides, ridges, and channels, respectively.
FIG. 2 is the intersection of skewness and kurtosis of the depth distribution of 5 kinds of landforms. The skewness of the depth distribution of the 5 landforms in the embodiment is larger than zero, the kurtosis of all terrace landforms and raised landforms is smaller than zero, the kurtosis of all gully landforms is larger than zero, and the kurtosis of a part of landslides and water channel landforms is larger than zero. Therefore, neither the terrace nor the raised relief in this embodiment has non-gaussian properties; all gullies, a part of landslides and water channel landforms have non-Gaussian properties, and multi-fractal spectrum characteristic values of the gullies and the landslides can be extracted.
(2) Calculating the characteristics of the multi-fractal spectrum:
according to respective Δ α ═ αmax-αmin、f(α(q))max=max(f(α(q)))、Δf=f(αmax)-f(αmin) And DqCalculating fractal spectrum width Δ α, fractal spectrum peak f (α (q))maxMaximum and minimum fractal dimension difference delta f and dimension DqWherein, when q is 0, DqFor capacity dimension, q is 2 hours DqFor the correlation dimension, max (f (α (q))) represents taking the maximum value of f (α (q)), α (q) being the Hausdor dimension of the topographic depth distribution of the area to be identified, αmaxIs the maximum singularity index, alphaminF (alpha (q)) is a distribution density function, alphamax、αminAlpha (q) and f (alpha (q)) are calculated by using a Chhabra algorithm, and q is a scale parameter and is set in the Chhabra algorithm.
Fig. 3 is a cross-sectional view of the multi-fractal spectral features of a non-gaussian landform in this embodiment. The related multi-fractal spectrum characteristics comprise multi-fractal spectrum width, multi-fractal spectrum peak value, maximum and minimum fractal dimension difference, capacity dimension and correlation dimension.
(3) Extracting landform factors based on multi-fractal spectrum characteristics:
original variables are Δ α, f (α (q))
maxΔ f and D
qExtracting a common factor f of the original variables by using a factor analysis method
iI is 1,2,3, … …, the variance contribution ratio of each factor is λ
iI is 1,2,3, … …, if satisfied
F is then
iI is 1,2, … … is a landform factor based on the characteristic of multi-fractal spectrum, wherein t is the number of common factors, f is
i、λ
iT is determined by factor analysis, theta is a threshold value set in the program parameters, and j is the number of topographic factors.
In this embodiment, the number of common factors of the original variables of the non-gaussian landform is t ═ 5, fig. 4 shows the variance contribution rate distribution of 5 common factors of the 5 multi-fractal spectral features of the non-gaussian landform type, and fig. 5 shows the cumulative variance contribution rate of 5 common factors of the 5 multi-fractal spectral features of the non-gaussian landform type. In this example, θ is 50%, and 2 topographic factors are obtained.
(4) And (3) designing a landform type classifier:
according to the landform factor fiAnd i is 1,2 and … …, and a support vector machine is applied to determine the landform type classifier.
In the present embodiment, 2 geomorphic factors are extracted, and fig. 6 shows a non-gaussian geomorphic type classifier in the present embodiment.
(5) And (3) landform type identification:
calculating the skewness and kurtosis value m of the terrain depth of the landform to be identified1And m2If m is satisfied1> 0 and m2If the number is more than 0, calculating the multi-fractal spectrum characteristics of the landform to be identified and the landform factor value f thereofiAnd (5) determining the type of the landform to be identified according to the classifier obtained in the step (4), wherein i is 1,2 and … ….
FIG. 7 is a diagram illustrating the recognition result of non-Gaussian landform types in this embodiment.