CN-121982302-A - Image segmentation method and system applied to marine plankton

CN121982302ACN 121982302 ACN121982302 ACN 121982302ACN-121982302-A

Abstract

The invention relates to the technical field of image data processing, in particular to a method and a system for dividing an image of marine plankton. The method comprises the steps of obtaining an in-situ microscopic image of marine plankton, carrying out binarization processing and connected domain analysis on the in-situ microscopic image, screening out a biological adhesion block to be processed, synchronously extracting edge contours of the biological adhesion block to be processed, calculating a biological structural integrity score, selecting a candidate segmentation path with the highest biological structural integrity score as an optimal segmentation line, carrying out curvature smooth complementation along the optimal segmentation line to obtain a single biological image, and inputting the single biological image into a pre-trained classifier to obtain a biological abundance data set. According to the invention, the self-adaptive adhesion segmentation is performed by screening and positioning adhesion lumps, and the classification of multidimensional morphological characteristics is combined, so that the high-precision automatic identification and abundance statistics of plankton in a complex marine environment are realized.

Inventors

WANG XIAO
SUN PING
LI BAOSHI
LIU PING
LI YAN
PU XINMING

Assignees

自然资源部第一海洋研究所

Dates

Publication Date: 20260505
Application Date: 20260106

Claims (10)

1. The method for segmenting the marine plankton image is characterized by comprising the following steps of: S1, acquiring an in-situ microscopic image of marine plankton, performing binarization processing and connected domain analysis on the in-situ microscopic image, screening a target area with the area exceeding a preset monomer area threshold, and marking the target area as a biological adhesion mass to be processed; S2, synchronously extracting edge contours of the biological adhesion blocks to be processed to locate contour concave extreme points and constructing a topological structure diagram; constructing a plurality of candidate segmentation paths based on the contour concave extreme point set, mapping the candidate segmentation paths into a topological structure diagram, and calculating a biological structure integrity score by combining the topological fracture cost of the candidate segmentation paths; s3, selecting a candidate segmentation path with the highest biological structure integrity score as an optimal segmentation line; S4, separating the biological adhesion block to be processed into independent subareas along an optimal separation line, and carrying out curvature smooth complementation on the edge of the separated incision to obtain a single biological image; and S5, inputting the single biological image into a pre-trained classifier to obtain a biological abundance data set containing species types, individual numbers and volume concentrations.
2. The method of claim 1, wherein the step S2 of synchronously extracting the edge contour of the biological adhesion mass to be processed to locate the contour concave extreme point comprises: Moving along the edge contour of the biological adhesion block to be processed through a preset sliding sampling window, and calculating covariance matrixes of coordinates of all pixel points in the window; Determining a discrete curvature value of a window center pixel point according to the eigenvalue ratio of the covariance matrix, and generating a full-contour curvature distribution curve; Searching extremum of the full-profile curvature distribution curve, identifying the trough position of which the curvature value is lower than that of the adjacent pixel point, and defining the trough position as a local curvature minimum value point; And setting a negative curvature threshold value, eliminating fluctuation with curvature value larger than the negative curvature threshold value from the local curvature minimum value points, and determining the rest local curvature minimum value points as contour concave extreme points.
3. The method for segmentation of marine plankton images according to claim 1, wherein the constructing of the topological structure diagram in step S2 includes: stripping the biological adhesion block mass to be treated layer by layer until a central skeleton line with single pixel width is reserved; Calculating the eight neighborhood communication numbers of pixel points on the central skeleton line, marking the points with the communication numbers larger than two as skeleton branch points, and marking the points with the communication numbers of one as skeleton end points; And (3) tracking connection paths among skeleton branch points, establishing a weighted undirected graph which takes skeleton branch points and skeleton endpoints as nodes and takes the connection paths as edges, and taking the pixel length of the connection paths as the edge weight of the weighted undirected graph to form a topological structure diagram.
4. The method of claim 1, wherein constructing a plurality of candidate segmentation paths based on the set of contour concave extremum points in step S2 comprises: calculating Euclidean distances among all contour concave extreme points, and pairing non-adjacent contour concave extreme points with the distances smaller than a preset adhesion width to form candidate segmentation point pairs; The candidate segmentation point pairs are connected to form a straight line segment, or the two segmentation point pairs extend along the normal direction of the two segmentation point pairs to form a folding line segment, and the straight line segment or the folding line segment is defined as a candidate segmentation path.
5. The method of claim 4, further comprising a validation of the pairing prior to defining the straight line segment or the broken line segment as the candidate segmentation path: Searching the pixel points with the nearest Euclidean distance on the central skeleton line of the topological structure diagram as skeleton projection points respectively by using two contour concave extreme points in the candidate segmentation point pairs; Searching the shortest skeleton path between two skeleton projection points in the topological structure diagram, and calculating the topological distance of the shortest skeleton path; And calculating the Euclidean distance between the two contour concave extreme points, if the ratio of the Euclidean distance to the topological distance is smaller than a preset path curvature threshold value, judging that the candidate segmentation point pair is effective, allowing to generate a candidate segmentation path, and otherwise, judging that the candidate segmentation path pair is ineffective.
6. The method of claim 1, wherein mapping the candidate segmented paths into the topological structure map in step S2, and calculating the biostructure integrity score in combination with the topological fracture costs of the candidate segmented paths comprises: calculating a concave depth-to-width ratio characteristic based on contour concave extreme points corresponding to two ends of the candidate segmentation path; Detecting whether the candidate segmentation paths are intersected with the connection paths in the topological structure diagram, and marking the skeleton edge of each intersection position as a topological interference edge; Assigning a topological interference coefficient according to the node types at two ends of the topological interference edge; and obtaining the original edge weight of the topological interference edge from the topological structure diagram, and calculating the biological structural integrity score of the candidate segmentation path according to the concave depth-to-width ratio characteristic, the original edge weight and the topological interference coefficient.
7. The method for segmentation of marine plankton images according to claim 1, wherein step S4 comprises the steps of: s41, determining two intersection points of an optimal dividing line and edge contours in the biological adhesion block mass to be processed as anchor points for incision repair; S42, extracting tangential slopes of two anchor points in the contour neighborhood of the biological adhesion block to be processed, and taking the tangential slopes as boundary derivative constraint conditions of incision repair; Step S43 of generating a smooth transition curve between the two anchor points in combination with the boundary derivative constraint, And S44, replacing the linear optimal dividing line with the smooth transition curve to form a closed monomer biological edge, thereby obtaining a monomer biological image.
8. The method for segmentation of marine plankton images according to claim 7, wherein step S4 further comprises the steps of: step S45, extracting the subarea area and the convex hull filling rate of the single biological image; step S46, comparing the area of the subareas with a preset single area threshold value, and comparing the filling rate of the convex hulls with a preset compactness threshold value; Step S47, if the area of the subarea is larger than a preset single area threshold value and the filling rate of the convex hull is lower than a preset compactness threshold value, judging the single biological image as a residual adhesion block mass; and S48, taking the residual adhesion block mass as a new biological adhesion block mass to be processed, and returning to the steps S2 to S4 until all the output monomer biological images do not meet the judgment condition of the step S47.
9. The method for segmentation of marine plankton images according to claim 1, wherein step S5 comprises the steps of: S51, extracting a morphological characterization vector of the single biological image; S52, inputting the morphological characterization vector into a pre-trained plankton classifier, identifying the specific species category of each monomer organism, and giving a species label; Step S53, counting the number of labels of the same species, and calculating the volume concentration of each species by combining the sampling volume in the in-situ microscopic image to generate a biological abundance data set containing species types, individual numbers and volume concentrations.
10. An application to a marine plankton image segmentation system for performing the application to a marine plankton image segmentation method as set forth in claim 1, the application to a marine plankton image segmentation system comprising: The adhesion block screening module is used for acquiring an in-situ microscopic image of marine plankton, carrying out binarization processing and connected domain analysis on the in-situ microscopic image, screening a target area with the area exceeding a preset monomer area threshold value, and marking the target area as a biological adhesion block to be processed; the system comprises a structural integrity analysis module, a topological structure graph, a biological structural integrity score calculation module, a biological structural integrity analysis module and a biological structural integrity analysis module, wherein the structural integrity analysis module is used for synchronously extracting the edge contour of a biological adhesion block mass to be processed so as to position a contour concave extreme point and construct a topological structure graph; The optimal segmentation decision module is used for selecting a candidate segmentation path with the highest biological structure integrity score as an optimal segmentation line; the monomer separation repair module is used for separating the biological adhesion block to be processed into independent subregions along the optimal separation line, and carrying out curvature smooth complementation on the edge of the separated incision to obtain a monomer biological image; The biological classification statistical module is used for inputting the single biological image into a pre-trained classifier to obtain a biological abundance data set containing species category, individual number and volume concentration.

Description

Image segmentation method and system applied to marine plankton Technical Field The invention relates to the technical field of image data processing, in particular to a method and a system for dividing an image of marine plankton. Background In the actual in-situ imaging process, plankton is always in a three-dimensional swimming state, serious overlapping and adhesion can be formed in red tide or high-abundance water areas, meanwhile, different species have great morphological differences, and many plankton carry slender antennae, bristles or complex limb protrusions. These factors act together to cause blurring of the edges and aliasing of textures of the target in the image, which can easily confuse natural morphological boundaries and adhesion boundaries of biological individuals, and ultimately affect the accuracy of the biological abundance statistics. The traditional watershed algorithm or the segmentation strategy based on simple geometric pits, which is commonly applied to the marine plankton image segmentation technology, has the logical defects caused by incapability of distinguishing 'biological limb connection' and 'inter-individual adhesion', and is easy to misjudge complex protrusion structures (such as antennae and bristle roots) of organisms as adhesion boundaries, so that excessive segmentation of the organisms is caused, a large number of fragmented pseudo targets are generated, and the final biological abundance statistical data is seriously distorted. Disclosure of Invention Based on the above, the present invention provides a method and a system for dividing images of marine plankton, so as to solve at least one of the above technical problems. In order to achieve the above object, a method for segmenting an image of marine plankton comprises the following steps: S1, acquiring an in-situ microscopic image of marine plankton, performing binarization processing and connected domain analysis on the in-situ microscopic image, screening a target area with the area exceeding a preset monomer area threshold, and marking the target area as a biological adhesion mass to be processed; S2, synchronously extracting edge contours of the biological adhesion blocks to be processed to locate contour concave extreme points and constructing a topological structure diagram; constructing a plurality of candidate segmentation paths based on the contour concave extreme point set, mapping the candidate segmentation paths into a topological structure diagram, and calculating a biological structure integrity score by combining the topological fracture cost of the candidate segmentation paths; s3, selecting a candidate segmentation path with the highest biological structure integrity score as an optimal segmentation line; S4, separating the biological adhesion block to be processed into independent subareas along an optimal separation line, and carrying out curvature smooth complementation on the edge of the separated incision to obtain a single biological image; and S5, inputting the single biological image into a pre-trained classifier to obtain a biological abundance data set containing species types, individual numbers and volume concentrations. The present invention also provides a method for applying to a marine plankton image segmentation system, the method comprising: The adhesion block screening module is used for acquiring an in-situ microscopic image of marine plankton, carrying out binarization processing and connected domain analysis on the in-situ microscopic image, screening a target area with the area exceeding a preset monomer area threshold value, and marking the target area as a biological adhesion block to be processed; the system comprises a structural integrity analysis module, a topological structure graph, a biological structural integrity score calculation module, a biological structural integrity analysis module and a biological structural integrity analysis module, wherein the structural integrity analysis module is used for synchronously extracting the edge contour of a biological adhesion block mass to be processed so as to position a contour concave extreme point and construct a topological structure graph; The optimal segmentation decision module is used for selecting a candidate segmentation path with the highest biological structure integrity score as an optimal segmentation line; the monomer separation repair module is used for separating the biological adhesion block to be processed into independent subregions along the optimal separation line, and carrying out curvature smooth complementation on the edge of the separated incision to obtain a monomer biological image; The biological classification statistical module is used for inputting the single biological image into a pre-trained classifier to obtain a biological abundance data set containing species category, individual number and volume concentration. The beneficial effects of the invention are as follows: The edge cont