Molecular cloud core detection method based on Facet
Through the molecular cloud nuclear detection method combined with the Facet model and the multivariate function maximum value theory, the problem of parameter sensitivity and manual judgment dependence in the existing technology is solved, and stronger noise resistance and automated detection are achieved, which is in line with the star formation process. Actual situation.
Patent Information
- Application Number
- CN202210134770.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-14
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2042-02-14
AI Technical Summary
The existing molecular cloud nuclear detection algorithm is sensitive to parameters and relies on manual judgment. The detection type is not easy to generalize, and its performance is reduced in a different signal-to-noise ratio environment. The detection results are greatly affected by subjective factors and do not conform to the actual situation of the star formation process. .
采用基于Facet的分子云核检测方法,通过构建三维高斯分布模型,进行形态学预处理,使用多元函数极大值理论和Facet模型进行中心检测,结合梯度分割和连通性最小距离聚类,实现分子云核的自动化检测。
It improves the noise resistance of detection, reduces dependence on parameters, improves the degree of automation of detection, the detection results are more in line with the actual situation of star formation, and are more stable in complex astronomical data.
Smart Images

Figure CN114724018B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of molecular cloud nucleus detection, and in particular to a molecular cloud nucleus detection method based on Facet. Background Art
[0002] Molecular gas is an important component of the interstellar medium and plays an important role in the evolution of the universe. Understanding the origin of the stellar initial mass function is one of the foundations for unlocking the mystery of stellar evolution. Automatic and accurate detection of molecular cloud cores is an urgent technical requirement for large-scale sky measurement analysis.
[0003] Molecular cloud cores have local high density and fluid characteristics. For the detection of molecular cloud cores, the algorithms involved in the prior art are: GaussClumps, ClumpFind, Reinhold, FellWalker, LDC and ConBased algorithms. GaussClumps works well for Gaussian distributed cluster detection, but there are a large number of irregularly distributed clusters in the observation data. ClumpFind is very sensitive to the specific value used for contour intervals; Reinhold is more susceptible to noise because it uses lower pixel values to find block edges; FellWalker and LDC are sensitive to the maximum jump distance, which is used to determine the range of extreme points that need to be merged, and LDC also depends on the minimum peak intensity; although ConBased takes into account the three physical parameters of volume, intensity difference and center distance difference, the influence of the parameters is still large. These algorithms significantly reduce the detection performance in an environment with a poor signal-to-noise ratio and detect the location of blocks with large deviations. GaussClumps lacks regional information of molecular cloud cores, while other algorithms have poor regional segmentation accuracy. FellWalker and LDC can segment the detected molecular cloud cores into discrete parts. Molecular clouds themselves have complex unknown conditions, and the detection results of these algorithms are greatly affected by subjective factors. Summary of the invention
[0004] In order to solve the defects of the existing molecular cloud core detection technology, such as sensitivity to parameters, great reliance on manual judgment, difficulty in generalizing the detection type, and inconsistency with the known star formation process, the present invention provides a molecular cloud core detection method based on Facet, which has strong anti-noise ability, low reliance on parameters, better comprehensive effect of target detection, and the detected molecular cloud core is more consistent with the current understanding of star formation.
[0005] The technical solution adopted by the present invention is:
[0006] The molecular cloud nucleus detection method based on Facet includes the following steps:
[0007] Step 1: Construct a molecular cloud core model;
[0008] Step 2: Perform morphological preprocessing on the original data to extract the signal area;
[0009] Step 3: Detect the molecular cloud core center in the signal area based on the multivariate function maximum theory and the Facet model;
[0010] Step 4: Perform gradient-based segmentation on the signal area to obtain the local area;
[0011] Step 5: Clustering the local area as the center of the molecular cloud core based on the minimum distance clustering of connectivity;
[0012] Through the above steps, the detection of the molecular cloud nucleus is achieved.
[0013] In step 1, the molecular cloud core model is a three-dimensional Gaussian distribution model, and its distribution function is:
[0014]
[0015] Where: Σ represents the variable The covariance matrix of represents the mean, A represents the amplitude, n represents the dimension of the Gaussian distribution, and T represents the transpose.
[0016] The molecular cloud core model can be rotated at any angle along the third velocity axis. The rotation equation is:
[0017] x′=(x-x0)cos(θ)-(y-y0)sin(θ)+x0
[0018] y′=(x-x0)sin(θ)+(y-y0)cos(θ)+y0
[0019] Where: (x0, y0) is the center coordinate of the cloud core on the xy plane, x, y are the coordinates of the molecular cloud core on the xy plane before rotation, x', y' are the coordinates of the molecular cloud core on the xy plane after rotation, and θ is the rotation angle; the cloud core can be truncated according to needs. The present invention defaults to truncation with the half-height weight width of the Gaussian distribution The truncation condition equation is:
[0020]
[0021] Where: σ is the diagonal element of the covariance matrix, x i ,y j ,z k Represent the i, j, and k coordinates of the Gaussian distribution on the x, y, and z axes respectively.
[0022] The above three-dimensional Gaussian distribution model is used to construct a single molecular cloud core. The data block is composed of multiple simulated cloud cores. The present invention requires that when adding a single cloud core, it is necessary to judge the change in the number of peaks in the data block after adding the single cloud core. There are three situations: the peak value remains unchanged, the peak value increases by one, and the peak value increases by multiple. Only the cloud core that increases the number of peaks in the data block by one can be added to the data block, and the divisibility standard of the molecular cloud core is determined by this. The peak point is defined as a point within a 3×3×3 neighborhood centered on any point in the data block, where the intensity of any point except the center point itself is smaller than its intensity.
[0023] The step 2 comprises the following steps:
[0024] Step 2.1, threshold segmentation:
[0025] Select a threshold to binarize the original data, and mark the pixel values greater than the threshold as 1. The data marked as 1 is the binary mask that needs to continue morphological processing;
[0026] Step 2.2, open operation:
[0027] Perform morphological opening operation on the binary mask to obtain the opening operation mask. The opening operation can separate the boundary of the signal area from the noise, thereby eliminating the noise points, and has no significant effect on the size of the signal area;
[0028] Step 2.3, expansion:
[0029] In order to allow more useful signals to participate in the detection, the opening operation mask is morphologically expanded to obtain the expanded mask. When the expansion area is expanded, intensity values below the noise level may be introduced. In order to remove these values below the threshold, the intersection of the binary mask and the expanded mask is extracted, which is called the intersection mask;
[0030] Step 2.4, connected domain marking:
[0031] Apply the connectivity operator to the intersection mask to obtain the connected domain, and multiply the connected domain with the original data to obtain the signal region.
[0032] In step 3, based on the maximum value theory of multivariate function, the specific steps are as follows:
[0033] The Hessian Matrix is a square matrix composed of the second-order partial derivatives of a multivariate function, which describes the local curvature of the function. The Hessian Matrix is often used in Newton's method to solve optimization problems. The Hessian Matrix can be used to determine the extreme value problem of a multivariate function.
[0034] Suppose n multivariate real functions f(x1,x2,...,x n ) at point M0(t1,t2,...,t n) has a second-order continuous partial derivative in the neighborhood of x1,x2,...,x n Represents the unknowns of a multivariate real function, t1, t2, ..., t n Represents the coordinates of M0.
[0035] If yes: and:
[0036]
[0037] When A is a negative definite matrix, f(x1,x2,...,x n ) in M0(t1,t2,...,t n ) is a maximum. The necessary and sufficient condition for a matrix to be a negative definite matrix is that all its eigenvalues are less than zero.
[0038] In step 3, the Facet model is as follows:
[0039] The Facet model uses the idea of segmentation to find the analytical function that best approximates the intensity value of the neighborhood in a regular area centered on a pixel using the least squares fitting method. For simple images, a flat model, an oblique model, or a quadratic model can be used. For images in complex areas, a cubic model or a higher-order model is used.
[0040] The ternary cubic polynomial of the Facet model used in the present invention is:
[0041] f(x,y,z)=a1g1(x,y,z)+a2g2(x,y,z)+...+a 20 g 20 (x,y,z) (1);
[0042] Among them, a1, a2, ..., a 20 represents the fitting coefficients, g1, g2, ..., g 20 represents basis function;
[0043] Right now:
[0044] in,
[0045] Φ=(g1,g2,...,g N )={1,x,y,z,x 2 ,y 2 ,z 2 ,xy,xz,yz,xyz,xy 2 ,xz 2 ,x 2 y,yz 2 ,x 2 z,y2 z,x 3 ,y 3 ,z 3 );
[0046] coefficient The objective function ε is minimized by least squares fitting.
[0047]
[0048] In formula (3), I(x, y, z) is the original surface, f(x, y, z) is the fitted surface, and W(x, y, z) is the Gaussian window function, that is:
[0049]
[0050] Where s is the window scale. The function space spanned by the basis function Φ is defined as:
[0051]
[0052] Among them, p(x,y,z),q(x,y,z) are arbitrary ternary functions, and W(x,y,z) is a Gaussian window function.
[0053] Then the error function ε in (3) can be written as:
[0054]
[0055] Where I is the original surface, Φ is the basis vector, is the fitting coefficient, W is the Gaussian window function, and T represents the transpose.
[0056] To differentiate (6), let The error function ε can be minimized, and we get:
[0057]
[0058] in, is the fitting coefficient, I is the original surface, Φ is the basis vector, K is the Facet model operator, and T represents the transpose. Substitute Φ into formula (7) and use formula (5) to calculate the weighted inner product to obtain the fitting coefficient: Using the 3DIDDG operator, the gradient vector G is obtained as:
[0059] G=(F x ,F y ,F z );
[0060] Among them: F x ,F y ,F zrepresent the partial derivatives of f(x,y,z) with respect to x,y,z respectively.
[0061]
[0062] Among them, a i is the fitting coefficient, and L is the 3D IDDG operator parameter.
[0063] When L = 0, the IDDG operator is simplified to a standard gradient operator that directly calculates partial derivatives. The 3D IDDG operator can effectively suppress noise. The continuous form of the inner product is not only more accurate, but also the window scale s is included in the final result. In order to accurately detect high-density cloud cores, the window scale s is set to 3 in the method of the present invention.
[0064] For f(x,y,z) (x,y,z)=(0,0,0) Find the second-order partial derivative:
[0065]
[0066] Among them, a i ,i=5,6,...,10 are the fitting coefficients.
[0067] Hessian matrix:
[0068]
[0069] Among them, A is the Hessian matrix, a i ,i=5,6,...,10 are the fitting coefficients.
[0070] The step 3 comprises the following steps:
[0071] Step 3.1: According to the theory of multivariate function maximum determination and considering the interference of noise, the first-order derivative of the center of the molecular cloud core is near zero, and the eigenvalue of the Hessian matrix of the center is less than zero. Therefore, when detecting the center, we do not directly look for the maximum value of the fitting surface, but extract the maximum value area within a certain range. However, there are large differences in the first-order derivatives and eigenvalues of different fitting surfaces. For example, the first-order derivative of the fitting surface in the flat area is small, while the first-order derivative in the steep area is large. In order to improve the robustness of the maximum area, an indirect method is used to obtain the adaptive threshold of the first-order derivative and eigenvalue of each signal area.
[0072] Step 3.1.1: Find the threshold of the first-order derivative:
[0073] The distribution of the first-order derivative can be fitted with a Gaussian function, whose mean is around 0. Therefore, the standard deviation of the first-order derivative values for each signal region is calculated, and the thresholds are positive double standard deviation and negative double standard deviation. The value coordinate set within the threshold is selected as the candidate value of the maximum region.
[0074] Step 3.1.2: For the threshold of the eigenvalue, the centers of different intensities have different eigenvalues. In order to adapt to the distribution characteristics of different signal areas, a coefficient K is given. bins , the number of intervals of eigenvalues will be obtained by equation (8). According to the Bins of each signal region, its eigenvalues are statistically analyzed. The threshold is the x-coordinate of the nth interval on the right. The initial value of n is 1, and the value increases by 1 after each recursion described later. The coordinate set that is less than the threshold is selected as the candidate value of the maximum region.
[0075]
[0076] Among them, Vsr i Represents the size of the i-th signal area, i=1,2,...,n, and n represents the number of signal areas.
[0077] Step 3.1.3: Extract the maximum value area:
[0078] According to the decision theorem, the relationship between the first-order derivative and the eigenvalue is a logical sum. In the pre-selected set, the position that meets the conditions is marked as 1. Then, the connected domain operator is applied to the marked data to mark the connected domain, and each connected domain is regarded as a maximum value area.
[0079] Step 3.2: After one screening, the obtained maximum region may still contain multiple molecular cloud core centers. Therefore, if the size of the maximum region is larger than Vmm, recursion is performed in the maximum region, and the judgment in step 3.1 is repeated to extract the sub-maximum region. In each recursion, the interval of the eigenvalue moves one grid to the left, and the selection method of the first-order derivative candidate remains unchanged.
[0080] Step 3.3: The available maximum region is defined as having a size less than Vmm and whose size multiplied by the number of recursions it corresponds to is greater than Vmm.
[0081] The available maximum region formulas are defined as follows:
[0082]
[0083] Where: Region i represents the ith maximum region, V i Represents Region i The size of R i Represents Regioni The number of recursions. All available maximum regions will be obtained in the above manner. The maximum region is combined with the signal region to obtain intensity information. Then the center of mass of each maximum region is calculated, and these centers of mass are recorded as the center of the molecular cloud core. The center of mass formula is defined as follows:
[0084]
[0085] Among them: Cen represents the centroid of a maximum value area, I represents the intensity distribution information, S represents the set of maximum region coordinates.
[0086] The step 4 is specifically as follows:
[0087] S4.1: Starting from any signal region, select a random point as the center of a box of size 3×3×3; then calculate the intensity gradient between it and each of its neighbors in the box. Move to the neighbor with the highest gradient and use it as the center of the next move. This can be described by the following equation:
[0088]
[0089] Where: L xyz represents the position of the next center, I represents the intensity distribution information, S represents the set of nearest neighbor coordinates in the box, u0 represents the midpoint in S, I[u i ] represents the coordinate u i The intensity value at the coordinate u0, I[u0] represents the intensity value at the coordinate u0, max(I[u i ]-I[u0]) represents the maximum intensity value, Represents the nearest neighbor coordinates corresponding to the maximum intensity value.
[0090] S4.2: Continue to move until the maximum gradient is less than zero. All points passed through during the movement are recorded as a path, and the end point of the path is recorded as a local maximum. When a local maximum is reached, a new local maximum will be searched from a new point that does not belong to any path until all points in all signal areas are searched. The points on the path that reach the same local maximum constitute a local area. At this point, all local maxima and local area information are obtained.
[0091] The step 5 depends on the signal region in step 2, the molecular cloud core center in step 3, the local region and the local maximum in step 4, and the goal is to make the local region gather to the center of each signal region.
[0092] The present invention provides a molecular cloud nucleus detection method based on Facet, and the technical effects are as follows:
[0093] 1) Each process of the detection method of the present invention has a strong theoretical basis, a high degree of automation, a small number of parameters, and low dependence on parameters.
[0094] 2) The detection method of the present invention has stronger anti-noise ability, smaller fluctuations in various performance curves, and better results.
[0095] 3) The detection method of the present invention is more adaptable to unknown and complex astronomical measured data and can be easily extended to other application fields. Taking into account the fluid characteristics of molecular cloud cores, the method of the present invention is more consistent with the understanding of the process of giant molecular clouds breaking up and forming long-lived low-mass stars in star formation. BRIEF DESCRIPTION OF THE DRAWINGS
[0096] Figure 1 It is a schematic diagram of the detection method of the present invention.
[0097] FIG2( a ) is a schematic diagram of the signal region of the 2D integration of simulation data;
[0098] Figure 2(b) is a schematic diagram of the signal area displayed in 3D of the simulation data;
[0099] Figure 2(c) is a schematic diagram of the signal region of the noise data.
[0100] Figure 2(d) is a schematic diagram of the signal region extracted based on graphics.
[0101] Figure 3 is the fitting surface of the Facet model.
[0102] Figure 4 This is the distribution information diagram of the initial first-order derivative of the signal area numbered 2.
[0103] Figure 5 This is the distribution information diagram of the initial eigenvalues of the signal area numbered 2.
[0104] FIG6( a ) is an integral diagram of the first recursive result of the signal region numbered 2;
[0105] FIG6( b ) is an integral diagram of the second recursive result of the signal region numbered 2;
[0106] FIG6( c ) is an integral diagram of the third recursive result of the signal region numbered 2;
[0107] FIG6( d ) is an integral diagram of the fourth recursive result of the signal region numbered 2. ...
[0108] Figure 7(a) is a schematic diagram of the final maximum area;
[0109] Figure 7(b) is a schematic diagram of the detected molecular cloud core center.
[0110] Figure 8Schematic diagram of local area segmentation results.
[0111] Fig. 9 Schematic diagram of the local area merging process.
[0112] Fig.10 Schematic diagram of the 2D integration of the molecular cloud nucleus detection results.
[0113] Fig.11 Schematic diagram of the 3D display of molecular cloud nucleus detection results.
[0114] Figure 12 (1) is a comparison chart of the recall rate R of simulation data detection;
[0115] Figure 12(2) is a comparison chart of the precision rate P of the simulation data detection;
[0116] Figure 12(3) is the F1 comparison chart of simulation data detection.
[0117] Figure 13 (1) is a comparison diagram of the center distance error △X detected by simulation data;
[0118] Figure 13 (2) is a comparison diagram of the flow error △flux detected by simulation data;
[0119] Figure 13(3) shows the comparison of the IOU of the regions detected by the simulation data.
[0120] Fig.14 Comparison chart of recall rate R for synthetic data detection.
[0121] Figure 15 (1) is a comparison of the center distance error △X detected by synthetic data on the Milky Way plane;
[0122] Figure 15(2) is a comparison chart of the center distance error △X of the speed axis detected by synthetic data.
[0123] Figure 16 (1) is a comparison diagram of the flow error △flux detected by synthetic data;
[0124] Figure 16(2) is a comparison chart of the IOU of regions detected by synthetic data.
[0125] Fig.17 The F1 of the cloud core with different signal-to-noise ratios varies with parameter K bins Graph of changes in . DETAILED DESCRIPTION
[0126] The method for detecting molecular cloud nuclei based on Facet, firstly, preprocesses the data based on morphology to extract the signal region. Then, the center of the molecular cloud nucleus is detected based on the maximum judgment theory of multivariate functions and Facet theory, and the 3DFacet model is used to fit the data to obtain the fitting information of the signal region; the Hessian matrix constructed by the first-order directional derivative of the three-dimensional IDDG operator and the second-order directional derivative of the fitting surface is used to extract the maximum region based on the extreme judgment theorem of multivariate functions. The centroid of the maximum region is taken as the center of the molecular cloud nucleus. At the same time, the signal region is segmented based on gradient to obtain the local region. Finally, the minimum distance clustering based on connectivity is performed to cluster the local region as the center of the molecular cloud nucleus to realize the detection of the molecular cloud nucleus. The following steps are included:
[0127] Step 1: Construct a molecular cloud core model;
[0128] In step 1, the molecular cloud core model is a three-dimensional Gaussian distribution model, and its distribution function is:
[0129]
[0130] Where: Σ represents the variable The covariance matrix of represents the mean, A represents the amplitude, n represents the dimension of the Gaussian distribution, and T represents the transpose.
[0131] The molecular cloud core model can be rotated at any angle along the third velocity axis (the first two axes are the galactic longitude and galactic latitude), and the rotation equation is:
[0132] x′=(x-x0)cos(θ)-(y-y0)sin(θ)+x0
[0133] y′=(x-x0)sin(θ)+(y-y0)cos(θ)+y0;
[0134] Where: (x0, y0) is the center coordinate of the cloud core on the xy plane, x, y are the coordinates of the molecular cloud core on the xy plane before rotation, x', y' are the coordinates of the molecular cloud core on the xy plane after rotation, θ is the rotation angle, and the cloud core can be truncated according to needs. The present invention defaults to truncation with the half-height weight width of the Gaussian distribution. The truncation condition equation is:
[0135]
[0136] Where: σ is the diagonal element of the covariance matrix, x i ,y j ,z k Represent the i, j, and k coordinates of the Gaussian distribution on the x, y, and z axes respectively.
[0137] The above three-dimensional Gaussian distribution model is used to construct a single molecular cloud core. The data block is composed of multiple simulated cloud cores. The present invention requires that when adding a single cloud core, it is necessary to judge the change in the number of peaks in the data block after adding the single cloud core. There are three situations: the peak value remains unchanged, the peak value increases by one, and the peak value increases by multiples. Only the cloud core that increases the number of peaks in the data block by one can be added to the data block, and this is used to determine the divisibility standard of the molecular cloud core. The peak point is defined as a point in the 3×3×3 neighborhood centered on any point in the data block, where the intensity of any point except the center point itself is smaller than its intensity. Step 2: Perform morphological preprocessing on the original data to extract the signal area;
[0138] The step 2 comprises the following steps:
[0139] Step 2.1, threshold segmentation:
[0140] Select a threshold to binarize the original data, and mark the pixel values greater than the threshold as 1. The data marked as 1 is the binary mask that needs to continue morphological processing;
[0141] Step 2.2, open operation:
[0142] Perform morphological opening operation on the binary mask to obtain the opening operation mask. The opening operation can separate the boundary of the signal area from the noise, thereby eliminating the noise points, and has no significant effect on the size of the signal area;
[0143] Step 2.3, expansion:
[0144] In order to allow more useful signals to participate in the detection, the opening operation mask is morphologically expanded to obtain the expanded mask. When the expansion area is expanded, intensity values below the noise level may be introduced. In order to remove these values below the threshold, the intersection of the binary mask and the expanded mask is extracted, which is called the intersection mask;
[0145] Step 2.4, connected domain marking:
[0146] Apply the connectivity operator to the intersection mask to obtain the connected domain, and multiply the connected domain with the original data to obtain the signal region.
[0147] The figure corresponding to step 2 is Figure 2(a) to Figure 2(d) ,Figure 2(d) shows the initial extracted signal region.
[0148] Step 3: Detect the molecular cloud core center in the signal area based on the multivariate function maximum theory and the Facet model;
[0149] In step 3, based on the maximum value theory of multivariate function, the specific steps are as follows:
[0150] The Hessian Matrix is a square matrix composed of the second-order partial derivatives of a multivariate function, which describes the local curvature of the function. The Hessian Matrix is often used in Newton's method to solve optimization problems. The Hessian Matrix can be used to determine the extreme value problem of a multivariate function.
[0151] Suppose n multivariate real functions f(x1,x2,...,x n ) at point M0(t1,t2,...,t n ) has a second-order continuous partial derivative in the neighborhood of x1,x2,...,x n Represents the unknowns of a multivariate real function, t1, t2, ..., t n Represents the coordinates of M0.
[0152] If yes: and:
[0153]
[0154] When A is a negative definite matrix, f(x1,x2,...,x n ) in M0(t1,t2,...,t n ) is a maximum. The necessary and sufficient condition for a matrix to be a negative definite matrix is that all its eigenvalues are less than zero.
[0155] In step 3, the Facet model is as follows:
[0156] The Facet model uses the idea of segmentation to find the analytical function that best approximates the intensity value of the neighborhood in a regular area centered on a pixel using the least squares fitting method. For simple images, a flat model, an oblique model, or a quadratic model can be used. For images in complex areas, a cubic model or a higher-order model is used.
[0157] The ternary cubic polynomial of the Facet model used in the present invention is:
[0158] f(x,y,z)=a1g1(x,y,z)+a2g2(x,y,z)+...+a 20 g 20 (x,y,z) (1);
[0159] Among them, a1, a2, ..., a 20 represents the fitting coefficients, g1, g2, ..., g 20 represents basis function;
[0160] Right now:
[0161] in,
[0162] Φ=(g1,g2,...,g N )={1,x,y,z,x 2 ,y 2 ,z 2 ,xy,xz,yz,xyz,xy 2 ,xz 2 ,x 2 y,yz 2 ,x 2 z,y 2 z,x 3 ,y 3 ,z 3 );
[0163] coefficient The objective function ε is minimized by least squares fitting.
[0164]
[0165] In formula (3), I(x, y, z) is the original surface, f(x, y, z) is the fitted surface, and W(x, y, z) is the Gaussian window function, that is:
[0166]
[0167] Where s is the window scale. The function space spanned by the basis function Φ is defined as:
[0168]
[0169] Among them, p(x,y,z),q(x,y,z) are arbitrary ternary functions, and W(x,y,z) is a Gaussian window function.
[0170] Then the error function ε in (3) can be written as:
[0171]
[0172] Where I is the original surface, Φ is the basis vector, is the fitting coefficient, W is the Gaussian window function, and T represents the transpose.
[0173] To differentiate (6), let The error function ε can be minimized, and we get:
[0174]
[0175] in, is the fitting coefficient, I is the original surface, Φ is the basis vector, K is the Facet model operator, and T represents the transpose. Substitute Φ into formula (7) and use formula (5) to calculate the weighted inner product to obtain the fitting coefficient: Using the 3DIDDG operator, the gradient vector G is obtained as:
[0176] G=(F x ,F y ,F z );
[0177] Among them: F x ,F y ,F z represent the partial derivatives of f(x,y,z) with respect to x,y,z respectively.
[0178]
[0179] Among them, a i is the fitting coefficient, and L is the 3D IDDG operator parameter.
[0180] L is the parameter of the 3D IDDG operator. When L = 0, the IDDG operator is simplified to a standard gradient operator that directly calculates partial derivatives. The 3D IDDG operator can effectively suppress noise. The continuous form of the inner product is not only more accurate, but also the window scale s is included in the final result. In order to accurately detect high-density cloud cores, the window scale s is set to 3 in the method of the present invention.
[0181] For f(x,y,z) (x,y,z)=(0,0,0) Find the second-order partial derivative:
[0182]
[0183] Among them, a i ,i=5,6,...,10 are the fitting coefficients.
[0184] Hessian matrix:
[0185]
[0186] Among them, A is the Hessian matrix, a i ,i=5,6,...,10 are the fitting coefficients.
[0187] The step 3 comprises the following steps:
[0188] Step 3.1: According to the theory of maximum determination of multivariate functions and considering the interference of noise, the first-order derivative of the center of the molecular cloud core is near zero, and the eigenvalue of the Hessian matrix of the center is less than zero. Therefore, when detecting the center, we do not directly look for the maximum value of the fitting surface, but extract the maximum value area within a certain range. However, there are large differences in the first-order derivatives and eigenvalues of different fitting surfaces. For example, the first-order derivative of the fitting surface in the flat area is small, while the first-order derivative in the steep area is large. In order to improve the robustness of the maximum value area, an indirect method is used to obtain the adaptive threshold of the first-order derivative and eigenvalue of each signal area.
[0189] Step 3.1.1: Find the threshold of the first-order derivative. The distribution of the first-order derivative can be fitted by a Gaussian function, whose mean is around 0. Therefore, we calculate the standard deviation of the first-order derivative value for each signal region, and the threshold is the positive double standard deviation and the negative double standard deviation. The value coordinate set within the threshold is selected as the candidate value of the maximum region.
[0190] Step 3.1.2: For the threshold of the eigenvalue. Centers with different intensities have different eigenvalues. In order to adapt to the distribution characteristics of different signal areas, we give a coefficient K bins , the number of intervals of eigenvalues will be obtained by equation (8). According to the Bins of each signal region, its eigenvalues are statistically analyzed. The threshold is the x-coordinate of the nth interval on the right. The initial value of n is 1, and the value increases by 1 after each recursion described later. The coordinate set that is less than the threshold is selected as the candidate value of the maximum region.
[0191]
[0192] Among them, Vsr i Represents the size of the i-th signal area, i=1,2,...,n, and n represents the number of signal areas.
[0193] Step 3.1.3: Extract the maximum value region. According to the decision theorem, the relationship between the first-order derivative and the eigenvalue is a logical sum. In the pre-selected set, the position that meets the conditions is marked as 1. Then, the connected domain operator is applied to the marked data to mark the connected domain, and each connected domain is regarded as a maximum value region.
[0194] The figure corresponding to the above step 3.1 is Figure 4 and Figure 5 , Figure 4 Shows the first derivative F in the x direction x Distribution statistics, Gaussian fitting curve and selection method of candidate values, first-order derivative F y 、F z Information and F x resemblance; Figure 5The distribution statistics of the first eigenvalue and the selection method of candidate values are shown. The information of the second and third eigenvalues is similar to that of the first eigenvalue.
[0195] Step 3.2: After one screening, the obtained maximum region may still contain multiple molecular cloud core centers. Therefore, if the size of the maximum region is larger than Vmm, recursion is performed in the maximum region, and the three sub-steps in step 3.1 are repeated to extract the sub-maximum region. In each recursion, the interval of the eigenvalue moves one grid to the left, and the selection method of the first-order derivative candidate remains unchanged.
[0196] The figure corresponding to step 3.2 is Figure 6(a) to Figure 6(d) ,Depend on Figure 6(a) to Figure 6(d) It can be seen that the overlapping cloud cores in the example are separated after 4 recursions.
[0197] Step 3.3: The available maximum region is defined as having a size less than Vmm and whose size multiplied by the number of recursions it corresponds to is greater than Vmm.
[0198] The available maximum region formulas are defined as follows:
[0199]
[0200] Where: Region i represents the ith maximum region, V i Represents Region i The size of R i Represents Region i The number of recursions. All available maximum regions will be obtained in the above manner. The maximum region is combined with the signal region to obtain intensity information. Then the center of mass of each maximum region is calculated, and these centers of mass are recorded as the center of the molecular cloud core. The center of mass formula is defined as follows:
[0201]
[0202] Among them: Cen represents the centroid of a maximum value area, I represents the intensity distribution information, S represents the set of maximum region coordinates.
[0203] The figures corresponding to step 3.2 are Figure 7(a) and Figure 7(b). Figure 7(a) shows all the final maximum areas. After screening, 5 simulated cloud cores are correctly detected. Figure 7(b) shows the centroid of the maximum area.
[0204] Step 4: Perform gradient-based segmentation on the signal area to obtain the local area; Step 4 is as follows:
[0205] S4.1: Starting from any signal region, select a random point as the center of a box of size 3×3×3; then calculate the intensity gradient between it and each of its neighbors in the box. Move to the point with the highest gradient as the center of the next movement. This can be described by the following equation:
[0206]
[0207] Where: L xyz represents the position of the next center, I represents the intensity distribution information, S represents the set of nearest neighbor coordinates in the box, u0 represents the midpoint in S, I[u i ] represents the coordinate u i The intensity value at the coordinate u0, I[u0] represents the intensity value at the coordinate u0, max(I[u i ]-I[u0]) represents the maximum intensity value, Represents the nearest neighbor coordinates corresponding to the maximum intensity value.
[0208] S4.2: Continue to move until the maximum gradient is less than zero. All points passed through during the movement are recorded as a path, and the end point of the path is recorded as a local maximum. When a local maximum is reached, a new local maximum will be searched from a new point that does not belong to any path until all points in all signal areas are searched. The points on the path that reach the same local maximum constitute a local area. At this point, all local maxima and local area information are obtained.
[0209] The figure corresponding to step 4 is Figure 8 , different colors represent different local areas.
[0210] Step 5: Assign the local area to the center of the molecular cloud core based on the minimum distance clustering of connectivity;
[0211] The step 5 depends on the signal region in step 2, the molecular cloud core center in step 3, the local region and the local maximum in step 4, and the goal is to make the local region gather to the center of each signal region.
[0212] The step 5 comprises the following steps:
[0213] Step 5.1: The molecular cloud core center in Figure 7(b) is matched with the signal region in Figure 2(d). If the match fails, the signal region and its local regions will be deleted. These failed signal regions are usually formed by noise. Region 4 in Figure 2(d) is a failed signal region.
[0214] Step 5.2: The molecular cloud core center in Figure 7(b) is matched with the local region in Figure (8). If the match is successful, the matched local region re-records the initial region of this center. The matched local region and its local maximum will be deleted from the record in step 4. If the match fails, it means that a local region may correspond to multiple centers; in order to select a more accurate center for the initial region, one of the failed center and the matched center that is closer to the local maximum of the initial region will be retained, and the other will be removed. After all matches are completed, each filtered center is associated with an initial region.
[0215] Step 5.3: The filtered local regions are clustered to the center of the filtered molecular cloud core. Connectivity is a necessary and sufficient condition for clustering. In the case of connectivity, the local regions are merged into the initial region whose center is closest to its local maximum. Different signal regions are not connected, so in order to improve efficiency, the following operations are performed in each signal region. The specific operations are as follows Fig. 9 As shown, when the records of the local maximum and local area in step 4 are not empty, a three-layer loop is introduced:
[0216] In the first layer, a center A is randomly selected, 26 nearby local maxima are searched in the signal area corresponding to the center, and these local maxima are arranged in ascending order;
[0217] In the second layer, a local maximum value B is selected from the sorted local maximum values in turn, and the N centers closest to the local maximum value B are searched. The initial value of N is 1, and the value of N is increased by 1 after each three-layer cycle.
[0218] The third layer selects a center C from the sorted centers and determines whether the initial area of C is connected to the local area of B. If so, the local area of B is merged into the initial area of C, and the initial area of C is updated. The current local maximum value B and its local area are deleted from the record; if not, the local maximum value B and its local area are retained and wait for the next judgment;
[0219] The final initial region is the region of the molecular cloud core. At this point, the center and region information of all molecular cloud cores has been obtained, and the detection is completed. The detection results are as follows: Fig.10 and Fig.11 As shown, Fig.10 shows the integral image of the detection mask, Fig.11 The spatial segmentation information of each cloud core is displayed, and different colors represent different cloud cores.
[0220] Example:
[0221] The relevant parameters involved in the present invention are as follows:
[0222] 1): The experimental data of the detection method of the present invention comes from the 3D Gaussian model simulation data and the synthetic data synthesized by M16 and Gaussian simulation cloud core. The size of the 3D data is 100*100*100 pixels, 100 cloud cores are placed in each data block, the peak range is [0.69, 4.6], and the standard deviation of the Gaussian distribution is [2, 4]. The size of the synthetic data is 68*181*361, the number of simulated cloud cores is 50, the peak range is [0.69, 15], and the standard deviation of the Gaussian distribution is [2, 4].
[0223] 2): Set the initial parameters of the algorithm {Threshold = 2*RMS, K bins =10, Vmm=64}
[0224] RMS is the noise level of the data, Threshold is the threshold, K bins The number of intervals of the eigenvalue histogram, Vmm is the minimum value of the product of the number of recursions and the volume of the maximum area.
[0225] 3): The core table of the detection data can be calculated through the molecular cloud core center and area information obtained in step 5. The core table includes the peak position, peak intensity, center of mass, volume, flow rate and axis length of the cloud core.
[0226] 4): The detection kernel table and the simulation kernel table are matched through the kernel table matching algorithm, and the accuracy, recall rate, F1, center distance error, flow error and area intersection ratio of the detection results can be calculated.
[0227] The detected core table includes the peak position, peak intensity, centroid, volume, flow rate and axial length of the detected cloud core;
[0228] The simulation core table includes the peak position, peak intensity, center of mass, volume, flow rate and axis length of the simulation cloud core.
[0229] The detection core table and the simulation core table are matched by the core table matching algorithm. If the Euclidean distance between the center of a detected cloud core and the center of a simulated cloud core is not greater than 2 pixels (the minimum standard deviation of the Gaussian distribution), the detection cloud core is matched with the simulated cloud core.
[0230] The evaluation indicators include: recall rate R, precision rate P, F1, center distance error △X, flow error △flux and area intersection over union ratio IOU. The calculation formula is as follows:
[0231]
[0232] Where: TP represents the number of correctly detected cloud cores, FP represents the number of incorrectly detected cloud cores, FN represents the number of missed cloud cores, N represents the number of matched cloud cores, Cen represents the center of detection, Region represents the regional coordinate information, and di represents the i-th correct cloud core in the detection, s i Represents the i-th correct cloud core in the simulation. The higher the recall rate, the higher the proportion of cloud cores found; the higher the accuracy rate, the higher the proportion of correctly detected cloud cores to detected cloud cores; the higher the F1, the better the overall performance of the algorithm in detection; the smaller the △X, the more accurate the center position detected by the algorithm; the higher the △flux, the smaller the flow loss detected by the algorithm; the higher the IOU, the more accurate the regional information of the molecular cloud core detected by the algorithm.
[0233] Through experiments on three-dimensional simulation data, in order to compare the performance of different algorithms with traditional algorithms under different signal-to-noise ratios, R, P, F1, △X, △flux and IOU are introduced, and the curves of the detection method of the present invention, GaussClumps, FellWalker, LDC and ConBased under these six indicators are plotted, as shown in Figure 2. Figure 12(1) to Figure 12(3) , Figure 13(1) to Figure 13(3) shown.
[0234] from Figure 12(1) to Figure 12(3) It can be seen that the recall rate of the Facet algorithm is always higher than that of the FellWalker algorithm and the LDC algorithm, higher than that of the ConBased algorithm at high signal-to-noise ratio, and higher than that of the GaussClumps algorithm at low signal-to-noise ratio; the precision and F1 of the Facet algorithm are higher than those of other algorithms at all signal-to-noise ratios, indicating that the Facet algorithm has stronger noise resistance;
[0235] from Figure 13(1) to Figure 13(3) It can be seen that the center distance error of the Facet algorithm is smaller, the flow loss is lower than other algorithms, and the IOU is higher than other algorithms. This shows that the Facet algorithm is superior to other algorithms in terms of region restoration and region segmentation;
[0236] The performance evaluation indicators in the synthetic data are R, △X, △flux and IOU. The detection comparison curve is as follows: Fig.14 , Figure 15(1), Figure 15(2) and Figure 16(1), Figure 16(2). Fig.14It can be seen that the recall rate of the Facet algorithm in synthetic data is higher than that of other algorithms, and the fluctuation of the curve is smaller, which means that Facet can accurately detect more molecular cloud nuclei in complex background data, and has stronger adaptability and stability in measured data. As can be seen from Figure 15(1) and Figure 15(2), the center distance error of the molecular cloud nucleus detected by the Facet algorithm is smaller than that of other algorithms in both the galactic plane and velocity directions, which means that the center position detected by Facet is more accurate. As can be seen from Figure 16(1) and Figure 16(2), the molecular cloud nucleus detected by the Facet algorithm has the least flow loss and a larger IOU, which means that Facet can detect the molecular cloud nucleus more completely in complex background data, and the regional information of the molecular cloud nucleus is more accurate. Fig.17 It can be seen that the mean fluctuation of F1 is small, indicating that the parameters have little effect on the detection results of molecular cloud nuclei.
Claims
1. Molecular cloud core detection method based on Facet, Features The following steps are involved: Step 1: Construct a molecular cloud core model; Step 2: Perform morphological preprocessing on the original data to extract the signal area; Step 3: Detect the center of the molecular cloud core in the signal area based on the multivariate function maximum theory and the Facet model; Step 4: Perform gradient-based segmentation on the signal area to obtain the local area; Step 5: Assign the local area to the center of the molecular cloud core based on the minimum distance clustering of connectivity; Through the above steps, the detection of the molecular cloud nucleus is achieved; In step 3, based on the maximum value theory of multivariate function, the specific steps are as follows: Suppose n multivariate real function f(x 1 ,x 2 ,...,x n ) at point M 0 (t 1 ,t 2 ,...,t n ) has a second-order continuous partial derivative in the neighborhood of x 1 ,x 2 ,...,x n represents the unknown number of the multivariate real function, t 1 ,t 2 ,...,t n Represents M 0 The coordinates of If yes: and: When A is a negative definite matrix, f(x 1 ,x 2 ,...,x n ) in M 0 (t 1 ,t 2 ,...,t n ) is the maximum value. The necessary and sufficient condition for a matrix to be a negative definite matrix is that all its eigenvalues are less than zero. In step 3, the ternary cubic polynomial of the Facet model is: f(x,y,z)=a 1 g 1 (x,y,z)+a 2 g 2 (x,y,z)+...+a 20 g 20 (x,y,z) (1); Among them, a 1 ,a 2 ,...,a 20 represents the fitting coefficient, g 1 ,g 2 ,...,g 20 represents basis function; Right now: in, Φ=(g 1 ,g 2 ,…,g N )={1,x,y,z,x 2 ,y 2 ,z 2 ,xy,xz,yz,xyz,xy 2 ,xz 2 ,x 2 y,yz 2 ,x 2 z,y 2 z,x 3 ,y 3 ,z 3 ): coefficient Minimize the objective function ε through least squares fitting; In formula (3), I(x, y, z) is the original surface, f(x, y, z) is the fitted surface, and W(x, y, z) is the Gaussian window function, that is: Where s is the window scale; the function space spanned by the basis function Φ is defined as: Among them, p(x,y,z),q(x,y,z) are arbitrary ternary functions, and W(x,y,z) is a Gaussian window function; Then the error function ε in (3) is written as: Where I is the original surface, Φ is the basis vector, is the fitting coefficient, W is the Gaussian window function, and T represents transposition; To differentiate (6), let The error function ε can be minimized, and we get: in, is the fitting coefficient, I is the original surface, Φ is the basis vector, K is the Facet model operator, and T represents the transpose; Substitute Φ into formula (7) and use formula (5) to calculate the weighted inner product to obtain the fitting coefficient: Using the 3DIDDG operator, the gradient vector G is obtained as: G=(F x ,F y ,F z ); Among them: F x ,F y ,F z Represent the partial derivatives of f(x,y,z) with respect to x,y,z respectively; Among them, a i is the fitting coefficient, L is the 3D IDDG operator parameter; When L = 0, the IDDG operator is simplified to the standard gradient operator that directly calculates partial derivatives; For f(x,y,z) (x,y,z)=(0,0,0) Find the second-order partial derivative: Among them, a i ,i=5,6,...,10 are fitting coefficients; Hessian matrix: Among them, A is the Hessian matrix, a i ,i=5,6,...,10 are the fitting coefficients.
2. According to the Facet-based molecular cloud nucleus detection method of claim 1, Features: In step 1, the molecular cloud core model is a three-dimensional Gaussian distribution model, and its distribution function is: Where: Σ represents the variable The covariance matrix of represents the mean, A represents the amplitude, n represents the dimension of Gaussian distribution, and T represents transpose; The molecular cloud core model can be rotated at any angle along the axis, and the rotation equation is: x′=(xx 0 )cos(θ)-(yy 0 )sin(θ)+x 0 and′=(xx 0 )sin(θ)+(yy 0 )cos(θ)+y 0 ; Where: (x 0 ,y 0 ) is the center coordinate of the cloud core on the xy plane, x, y are the coordinates of the molecular cloud core on the xy plane before rotation, x', y' are the coordinates of the molecular cloud core on the xy plane before rotation, and θ is the rotation angle; the cloud core can be truncated according to needs, and the default is to truncate it with the half-height weight width of the Gaussian distribution The truncation condition equation is: Where: σ is the diagonal element of the covariance matrix, x i ,y j ,z k Represent the i, j, and k coordinates of the Gaussian distribution on the x, y, and z axes respectively.
3. According to the Facet-based molecular cloud core detection method of claim 2, Features: The three-dimensional Gaussian distribution model is used to construct a single molecular cloud core. The data block is composed of multiple simulated cloud cores. When adding a single cloud core, it is necessary to judge the change in the number of peaks in the data block after the addition of the single cloud core. There are three situations: the peak value remains unchanged, the peak value increases by one, and the peak value increases by multiples. Only the cloud core that increases the number of peaks in the data block by one can be added to the data block, and this is used to determine the divisibility standard of the molecular cloud core; the peak point is defined as a point in the 3×3×3 neighborhood centered on any point in the data block, where the intensity of any point except the center point itself is smaller than its intensity.
4. According to the Facet-based molecular cloud core detection method of claim 1, Features: The step 2 comprises the following steps: Step 2.1, threshold segmentation: Select a threshold to binarize the original data, and mark the pixel values greater than the threshold as 1. The data marked as 1 is the binary mask that needs to continue morphological processing; Step 2.2, open operation: Perform morphological opening operation on the binary mask to obtain an opening operation mask; Step 2.3, expansion: The opening operation mask is morphologically dilated to obtain a dilated mask; when the dilation expands the area, intensity values below the noise level are introduced; in order to remove these values below the threshold, the intersection of the binary mask and the dilated mask is extracted, which is called the intersection mask; Step 2.4, connected domain marking: Apply the connectivity operator to the intersection mask to obtain the connected domain, and multiply the connected domain with the original data to obtain the signal region.
5. According to the Facet-based molecular cloud nucleus detection method of claim 1, Features: The step 3 comprises the following steps: Step 3.1: According to the theory of multivariate function maximum determination and considering the interference of noise, the first-order derivative of the center of the molecular cloud core is near zero, and the eigenvalue of the Hessian matrix of the center is less than zero; when detecting the center, do not directly find the maximum value of the fitting surface, but extract the maximum value area within a certain range; Step 3.2: After one screening, the obtained maximum region may still contain multiple molecular cloud core centers; therefore, if the size of the maximum region is larger than Vmm, recursion is performed in the maximum region, and the judgment in step 3.1 is repeated to extract the sub-maximum region; In each recursion, the interval of eigenvalues is shifted one grid to the left, and the method of selecting first-order derivative candidates remains unchanged; Step 3.3: Define the available maximum value region as having a size less than Vmm, and its size multiplied by the corresponding number of recursions is greater than Vmm; the formula for the available maximum value region is defined as follows: Where: Region i represents the ith maximum region, V i Represents Region i The size of R i Represents Region i The number of recursions; all available maximum regions will be obtained in the above manner; the maximum region is combined with the signal region to obtain intensity information; then the center of mass of each maximum region is calculated, and these centers of mass are recorded as the center of the molecular cloud core; the center of mass formula is defined as follows: Among them: Cen represents the centroid of a maximum value area, I represents the intensity distribution information, S represents the set of maximum region coordinates.
6. According to the Facet-based molecular cloud nucleus detection method of claim 5, Features: Step 3.1 includes the following steps: Step 3.1.1: Find the threshold of the first-order derivative: The distribution of the first-order derivative can be fitted with a Gaussian function, whose mean is around 0; the standard deviation of the first-order derivative value of each signal region is calculated, and the threshold is the positive double standard deviation and the negative double standard deviation; the value coordinate set within the threshold is selected as the candidate value of the maximum region; Step 3.1.2: For the threshold of the eigenvalue, the centers of different intensities have different eigenvalues. In order to adapt to the distribution characteristics of different signal areas, a coefficient K is given. bins , the number of intervals of eigenvalues will be obtained by formula (8); according to the Bins of each signal region, its eigenvalue is statistically analyzed; the threshold is the x coordinate of the nth interval on the right; the initial value of n is 1, and after each recursion described later, the value increases by 1; the coordinate set smaller than the threshold is selected as the candidate value of the maximum region; Among them, Vsr i represents the size of the i-th signal area, i = 1, 2, ..., n, n represents the number of signal areas; Step 3.1.3: Extract the maximum value area: The relationship between the first-order derivative and the eigenvalue is a logical sum; in the pre-selected set, the position that meets the conditions is marked as 1; then, the connected domain operator is applied to the marked data to mark the connected domain, and each connected domain is regarded as a maximum value area.
7. According to the Facet-based molecular cloud core detection method of claim 6, Features: The step 4 comprises the following steps: S4.1: Starting from any signal region, select a random point as the center of a box of size 3×3×3; then calculate the intensity gradient between it and each of its neighbors in the box; move to the neighbor with the highest gradient and use it as the center of the next move; this can be described by the following equation: Where: L xyz represents the position of the next center, I represents the intensity distribution information, S represents the set of nearest neighbor coordinates in the box, and u 0 represents the midpoint in S, I[u i ] represents the coordinate u i The intensity value at 0 ] represents the coordinate u 0 The intensity value at the point, max(I[u i ]-I[u 0 ]) represents the maximum intensity value, Represents the nearest neighbor coordinates corresponding to the maximum intensity value; S4.2: Continue moving until the maximum gradient is less than zero; all points passed during the movement are recorded as a path, and the end point of the path is recorded as a local maximum; when a local maximum is reached, a new local maximum is searched from a new point that does not belong to any path until all points in all signal areas are searched; points on the path that reach the same local maximum constitute a local area.
8. According to the Facet-based molecular cloud nucleus detection method of claim 7, Features: The step 5 comprises the following steps: Step 5.1: The molecular cloud core center is matched with the signal region. If the match fails, the signal region and its local region will be deleted; these failed signal regions are usually formed by noise; Step 5.2: The center of the molecular cloud core is matched with the local region. If the match is successful, the matched local region re-records the initial region of this center; the matched local region and its local maximum value will be deleted from the record in step 4; if the match fails, it means that a local region may correspond to multiple centers; in order to select a more accurate center for the initial region, one of the failed center and the matched center that is closer to the local maximum value of the initial region will be retained, and the other will be removed; after all matches are completed, each filtered center is associated with an initial region; Step 5.3: The filtered local areas are clustered to the center of the filtered molecular cloud core; connectivity is a necessary and sufficient condition for clustering. In the case of connectivity, the local areas are merged into the initial area whose center is closest to its local maximum; different signal areas are not connected, so in order to improve efficiency, the following operations are performed in each signal area; when the records of the local maximum and the local area in step 4 are not empty, a three-layer loop is introduced: In the first layer, a center A is randomly selected, 26 nearby local maxima are searched in the signal area corresponding to the center, and these local maxima are arranged in ascending order; In the second layer, a local maximum value B is selected from the sorted local maximum values in turn, and the N centers closest to the local maximum value B are searched. The initial value of N is 1, and the value of N is increased by 1 after each three-layer cycle. The third layer selects a center C from the sorted centers and determines whether the initial area of C is connected to the local area of B. If so, the local area of B is merged into the initial area of C and the initial area of C is updated. The current local maximum value B and its local area are deleted from the record. If not, the local maximum value B and its local area are retained and wait for the next judgment. The final initial region is the region of the molecular cloud core; at this point, the center and region information of all molecular cloud cores has been obtained, and the detection is completed.
Citation Information
Patent Citations
Convective monomer detection method based on extended maximum value transformation
CN110309802A
