An underwater coral reef ecological diversity identification method and system

Abstract

The present application discloses an underwater coral reef ecological diversity recognition method and system, which relates to the field of marine ecological analysis, and includes: obtaining multiple remote sensing images of the area to be recognized, and performing multiple divisions on the multiple remote sensing images according to a preset variety of time period division methods to obtain multiple groups of time series image sets; according to the multiple groups of time series image sets, combining with a time series feature model to obtain multiple time series feature markers; performing multiple divisions on the multiple remote sensing images according to a variety of grid division methods to obtain a first image set, a second image set and a third image set; according to the first image set, the second image set and the third image set, combining with a preset ecological feature dictionary to perform feature matching to obtain multiple first markers, multiple second markers and multiple third markers; according to the multiple first markers and the multiple third markers, correcting the multiple time series feature markers to obtain multiple corrected feature markers, and according to the multiple corrected feature markers and the multiple second markers, obtaining the coral reef ecological recognition result of the area to be recognized.

Description

Technical Field

[0001] The present application relates to the field of marine ecological analysis, and in particular to a method and system for identifying underwater coral reef ecological diversity. Background Art

[0002] Coral reefs are one of the most representative ecosystems in tropical oceans. A rich variety of marine life lives near coral reef areas. Identifying and calculating the ecological populations of coral reefs is of great significance to the research and optimization of the marine environment and marine ecosystems.

[0003] In the prior art, when identifying the ecological populations of coral reefs, the identification is usually carried out based on remote sensing images or underwater real-time images of coral reefs and based on deep learning. However, it is generally only for short-term or real-time data, and the changes in ecological populations in the long term are not considered, resulting in differences in the precision and accuracy of identification. At the same time, when identifying, only the overall image is considered for identification, and the detailed identification of the local area is not considered, which will also lead to a decrease in the precision and accuracy of identification. Therefore, how to improve the identification precision and accuracy of the ecological populations of coral reefs is still a problem that needs to be solved urgently. Summary of the invention

[0004] The present application provides a method and system for identifying the ecological diversity of underwater coral reefs to solve the technical problem that the existing technology has insufficient precision and accuracy in identifying the ecological populations of coral reefs.

[0005] According to a first aspect of the embodiments of the present application, a method for identifying underwater coral reef ecological diversity is provided, comprising:

[0006] Acquire multiple remote sensing images of the area to be identified, and divide the multiple remote sensing images multiple times according to multiple preset time period division methods to obtain multiple groups of time series image sets; wherein the multiple time period division methods include 90 days, 180 days and 360 days;

[0007] According to the multiple groups of time-series image sets, combined with a preset time-series feature model, a plurality of time-series feature tags corresponding to the multiple remote sensing images are obtained;

[0008] The plurality of remote sensing images are divided multiple times according to a plurality of preset grid division methods to obtain a first image set, a second image set and a third image set; wherein the plurality of grid division methods are divided into one type of grid, two types of grid and three types of grid in descending order according to the size of the grid; the first image set is divided according to the one type of grid, the second image set is divided according to the two types of grid, and the third image set is divided according to the three types of grid;

[0009] Performing feature matching on the basis of the first image set, the second image set, and the third image set in combination with a preset ecological feature dictionary to obtain a plurality of first markers, a plurality of second markers, and a plurality of third markers;

[0010] Modifying the plurality of time-series feature markers according to the plurality of first markers and the plurality of third markers to obtain a plurality of modified feature markers, and obtaining a coral reef ecological recognition result of the area to be recognized according to the plurality of modified feature markers and the plurality of second markers.

[0011] In this application, first, remote sensing images are divided according to multiple time period division methods, and then a plurality of time-series feature markers are obtained in combination with a time-series feature model, which can extract the changes in the ecological population of coral reefs within different time periods, considering both the ecological changes within a short time range and those within a long time range, thereby improving the recognition accuracy; then, the remote sensing images are divided according to multiple grid division methods to obtain a first image set, a second image set, and a third image set, and feature matching is performed in combination with a preset ecological feature dictionary to obtain a plurality of first markers, a plurality of second markers, and a plurality of third markers, which can identify the ecological features of coral reefs at different granularities and embed the ecological features into the markers, achieving the purpose of identifying the coral reef ecological population; furthermore, the plurality of time-series feature markers are modified according to the plurality of first markers and the plurality of third markers, and a coral reef ecological recognition result to be recognized is obtained according to the plurality of modified feature markers and the plurality of second markers, which can combine the recognition results at different granularities and the recognition results with time-series features, further reducing the recognition error, thereby improving the recognition accuracy and accuracy of the coral reef ecological population.

[0012] In some embodiments of this application, the obtaining of multiple remote sensing images of the area to be recognized specifically includes:

[0013] Obtaining a plurality of initial remote sensing images and a plurality of initial infrared images of the area to be recognized;

[0014] Based on a preset feature recognition model, obtaining the common area features of the plurality of initial remote sensing images and the plurality of initial infrared images;

[0015] According to the common area features, performing information feature fusion on the plurality of initial remote sensing images and the plurality of initial infrared images to obtain a plurality of remote sensing images.

[0016] In this application, first, initial remote sensing images and initial infrared images are obtained, the common area features are obtained through a feature recognition model, and information feature fusion is performed on the initial remote sensing images and the initial infrared images based on the common area features to obtain remote sensing images, which can improve the information density in the remote sensing images, and thus is conducive to subsequent feature recognition and feature extraction, and is more adaptable to the actual needs of the task.

[0017] In certain embodiments of the present application, obtaining multiple temporal feature markers corresponding to the multiple remote sensing images by combining a preset temporal feature model according to the multiple sets of temporal image sets specifically includes:

[0018] Dividing each of the multiple sets of temporal image sets into multiple sets of temporal image training sets and corresponding multiple sets of temporal image prediction sets according to a preset training classification ratio;

[0019] Iteratively training the temporal feature model according to the multiple sets of temporal image training sets. In each round of iteration, input a set of temporal image training sets corresponding to the corresponding round into the temporal feature model for iteration to adjust the model parameters until the iteration round reaches the number of sets of the multiple sets of temporal image training sets, and obtain and output the first temporal feature model;

[0020] Input the multiple sets of temporal image prediction sets into the first temporal feature model to obtain multiple temporal feature markers corresponding to the multiple remote sensing images.

[0021] The present application first divides multiple sets of temporal image sets into training sets and prediction sets according to a preset training classification ratio, then iteratively trains the temporal feature model multiple times based on the training sets, and obtains multiple temporal feature markers corresponding to multiple remote sensing images based on the prediction sets, which can enable the temporal feature model to gradually learn the characteristics of the changes in the ecological populations of coral reefs, so as to be able to extract the changes in the ecological populations of coral reefs within different time periods and improve the recognition accuracy and precision.

[0022] In certain embodiments of the present application, performing multiple divisions on the multiple remote sensing images according to preset multiple grid division methods to obtain a first image set, a second image set, and a third image set specifically includes:

[0023] Dividing the multiple remote sensing images according to the grid size of the first type of grid to obtain a first grid image set;

[0024] Dividing the multiple remote sensing images according to the grid size of the second type of grid to obtain a second grid image set;

[0025] Dividing the multiple remote sensing images according to the grid size of the third type of grid to obtain an initial third grid image set, and performing feature enhancement on the initial third grid image set to obtain a third grid image set;

[0026] Performing denoising processing on the first grid image set, the second grid image set, and the third grid image set, and aligning the image feature information to obtain a first image set, a second image set, and a third image set.

[0027] In this application, multiple remote sensing images are first divided according to the grid sizes of the first type of grid and the second type of grid to obtain a first grid image set and a second grid image set. Then, the multiple remote sensing images are divided according to the grid size of the third type of grid and feature enhancement is performed to obtain a third grid image set, which can prevent the third grid image set divided by the grid size of the third type of grid from losing picture feature details. Furthermore, denoising is performed on the first grid image set, the second grid image set, and the third grid image set, and the image feature information is aligned, enabling the remote sensing images to be divided at different granularities, thereby meeting the subsequent requirement of distinguishing coral reef ecological population information at different granularities and being more adaptable to the actual task requirements.

[0028] In some embodiments of this application, the feature matching of the multiple first markers, multiple second markers, and multiple third markers is obtained by combining a preset ecological feature dictionary according to the first image set, the second image set, and the third image set, specifically including:

[0029] According to the first image set, the second image set, and the third image set, in combination with a preset feature extraction and encoding method, multiple first semantic features, multiple second semantic features, and multiple third semantic features are obtained;

[0030] Based on a similarity algorithm, feature matching is performed on the multiple first semantic features, the multiple second semantic features, and the multiple third semantic features according to a preset ecological feature dictionary to obtain multiple first markers, multiple second markers, and multiple third markers.

[0031] In some embodiments of this application, the feature extraction and encoding method specifically includes:

[0032] Feature extraction is performed on each image of the current image set to obtain multiple image feature vectors;

[0033] Semantic recognition and encoding are performed on the multiple image feature vectors to obtain multiple semantic features.

[0034] In this application, multiple first semantic features, multiple second semantic features, and multiple third semantic features are first extracted through a feature extraction and encoding method, and then based on a similarity algorithm, multiple first markers, multiple second markers, and multiple third markers are obtained through feature matching according to a preset ecological feature dictionary, which can identify the ecological features of coral reefs at different granularities and embed the ecological features into the markers, achieving the purpose of identifying coral reef ecological populations.

[0035] In some embodiments of this application, the multiple timing feature markers are corrected according to the multiple first markers and the multiple third markers to obtain multiple corrected feature markers, specifically including:

[0036] Eliminate the temporal feature markers that do not belong to the regions corresponding to the multiple first markers among the multiple temporal feature markers, to obtain multiple initial corrected feature markers;

[0037] According to the grid size of the type of grid corresponding to the multiple first markers, splice the multiple third markers to obtain multiple grid regions;

[0038] Eliminate the initial corrected feature markers that do not belong to the multiple grid regions among the multiple initial corrected feature markers, to obtain multiple corrected feature markers.

[0039] In this application, first, eliminate the temporal feature markers that do not belong to the regions corresponding to the multiple first markers among the multiple temporal feature markers to obtain multiple initial corrected feature markers, which can screen the temporal feature markers at the corresponding granularity of the first markers, and screen out some misidentified temporal feature markers. Then, according to the grid size of the type of grid corresponding to the first markers, splice the multiple third markers to obtain multiple grid regions, which can simplify the process of screening and eliminating through the third markers and improve the screening efficiency. Then, eliminate the initial corrected feature markers that do not belong to the multiple grid regions among the multiple initial corrected feature markers to obtain multiple corrected feature markers, which can screen the temporal feature markers at the corresponding granularity of the third markers and further screen out the misidentified items in the temporal feature markers, thereby improving the recognition accuracy and precision of the coral reef ecological population.

[0040] In some embodiments of this application, obtaining the coral reef ecological recognition result of the area to be recognized according to the multiple corrected feature markers and the multiple second markers specifically includes:

[0041] According to the multiple corrected feature markers and the multiple second markers, obtain a feature marker intersection and a feature marker symmetric difference set;

[0042] Screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold, to obtain a feature marker screening set;

[0043] Obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection and multiple second semantic recognition results corresponding to each feature marker in the feature marker screening set, and according to the multiple first semantic recognition results and the multiple second semantic recognition results, obtain the coral reef ecological recognition result of the area to be recognized.

[0044] In this application, first, an intersection of feature markers and a symmetric difference set of feature markers are obtained based on multiple correction feature markers and multiple second markers. Then, feature markers with recognition probabilities not greater than a preset recognition threshold in the symmetric difference set of feature markers are screened out to obtain a feature marker screening set, which can further exclude misrecognized feature markers. Furthermore, based on the semantic recognition results of the intersection of feature markers and the feature marker screening set, a coral reef ecological recognition result of the area to be recognized is obtained, which can integrate recognition results of different granularities and recognition results with temporal characteristics, and further improve the recognition accuracy and precision of the ecological population of coral reefs.

[0045] According to the second aspect of the embodiments of this application, an underwater coral reef ecological diversity recognition system is provided, including a time series division module, a time series marking module, a grid division module, a feature matching module, and a result acquisition module;

[0046] The time series division module is configured to obtain multiple remote sensing images of the area to be recognized, and perform multiple divisions on the multiple remote sensing images according to multiple preset time period division methods to obtain multiple groups of time series image sets; among them, the multiple time period division methods include 90 days, 180 days, and 360 days;

[0047] The time series marking module is configured to obtain multiple time series feature markers corresponding to the multiple remote sensing images according to the multiple groups of time series image sets in combination with a preset time series feature model;

[0048] The grid division module is configured to perform multiple divisions on the multiple remote sensing images according to multiple preset grid division methods to obtain a first image set, a second image set, and a third image set; among them, the multiple grid division methods are arranged in descending order of grid size and are divided into first-class grids, second-class grids, and third-class grids; the first image set is divided according to the first-class grids, the second image set is divided according to the second-class grids, and the third image set is divided according to the third-class grids;

[0049] The feature matching module is configured to perform feature matching on the first image set, the second image set, and the third image set in combination with a preset ecological feature dictionary to obtain multiple first markers, multiple second markers, and multiple third markers;

[0050] The result acquisition module is configured to correct the time series feature markers according to the multiple first markers and the multiple third markers to obtain multiple correction feature markers, and obtain a coral reef ecological recognition result of the area to be recognized according to the multiple correction feature markers and the multiple second markers.

[0051] In some embodiments of this application, the time series division module includes an image acquisition sub-module; the image acquisition sub-module includes an image acquisition unit, a feature recognition unit, and a feature fusion unit;

[0052] The image acquisition unit is configured to acquire multiple initial remote sensing images and multiple initial infrared images of the area to be recognized;

[0053] The feature recognition unit is configured to obtain the common area features of the multiple initial remote sensing images and the multiple initial infrared images based on a preset feature recognition model;

[0054] The feature fusion unit is configured to perform information feature fusion on the multiple initial remote sensing images and the multiple initial infrared images according to the common area features to obtain multiple remote sensing images.

[0055] In some embodiments of the present application, the timing marking module includes a data division sub-module, a model training sub-module, and a timing marking sub-module;

[0056] The data division sub-module is configured to divide each of the multiple groups of timing image sets into multiple groups of timing image training sets and corresponding multiple groups of timing image prediction sets according to a preset training classification ratio;

[0057] The model training sub-module is configured to perform iterative training on the timing feature model according to the multiple groups of timing image training sets. In each round of iteration, a group of timing image training sets corresponding to the current round is input into the timing feature model for iteration to adjust the model parameters until the number of iteration rounds reaches the number of groups of the multiple groups of timing image training sets, and then obtain and output a first timing feature model;

[0058] The timing marking sub-module is configured to input the multiple groups of timing image prediction sets into the first timing feature model to obtain multiple timing feature marks corresponding to the multiple remote sensing images.

[0059] In some embodiments of the present application, the grid division module includes a first division sub-module, a second division sub-module, a third division sub-module, and a feature processing sub-module;

[0060] The first division sub-module is configured to divide the multiple remote sensing images according to the grid size of the first type of grid to obtain a first grid image set;

[0061] The second division sub-module is configured to divide the multiple remote sensing images according to the grid size of the second type of grid to obtain a second grid image set;

[0062] The third division sub-module is configured to divide the multiple remote sensing images according to the grid size of the third type of grid to obtain an initial third grid image set, and perform feature enhancement on the initial third grid image set to obtain a third grid image set;

[0063] The feature processing submodule is used to perform denoising on the first grid image set, the second grid image set and the third grid image set, and align image feature information to obtain the first image set, the second image set and the third image set.

[0064] In certain embodiments of the present application, the feature matching module includes a feature extraction submodule and a feature matching submodule;

[0065] The feature extraction submodule is used to obtain a plurality of first semantic features, a plurality of second semantic features and a plurality of third semantic features according to the first image set, the second image set and the third image set in combination with a preset feature extraction encoding method;

[0066] The feature matching submodule is used to perform feature matching on the multiple first semantic features, the multiple second semantic features and the multiple third semantic features based on a similarity algorithm and a preset ecological feature dictionary to obtain multiple first tags, multiple second tags and multiple third tags.

[0067] In certain embodiments of the present application, the feature extraction encoding method specifically includes:

[0068] Extract features from each image in the current image set to obtain multiple image feature vectors;

[0069] Semantic recognition coding is performed on the multiple image feature vectors to obtain multiple semantic features.

[0070] In certain embodiments of the present application, the result acquisition module includes a feature correction submodule; the feature correction submodule includes an initial correction unit, a grid splicing unit and a feature correction unit;

[0071] The initial correction unit is used to remove the timing feature marks that do not belong to the regions corresponding to the multiple first marks from the multiple timing feature marks, so as to obtain multiple initial correction feature marks;

[0072] The grid splicing unit is used to splice the plurality of third marks to obtain a plurality of grid areas according to the grid size of the type of grids corresponding to the plurality of first marks;

[0073] The feature correction unit is used to eliminate the initial correction feature marks that do not belong to the multiple grid areas from the multiple initial correction feature marks to obtain multiple correction feature marks.

[0074] In certain embodiments of the present application, the result acquisition module includes a result acquisition submodule; the result acquisition submodule includes a marker aggregation unit, a marker screening unit and a result acquisition unit;

[0075] The marker aggregation unit is configured to obtain an intersection of feature markers and a symmetric difference set of feature markers according to the multiple corrected feature markers and the multiple second markers;

[0076] The marker screening unit is configured to screen out the feature markers in the symmetric difference set of feature markers whose corresponding recognition probabilities are not greater than a preset recognition threshold, so as to obtain a feature marker screening set;

[0077] The result acquisition unit is configured to obtain multiple first semantic recognition results corresponding to each feature marker in the intersection of feature markers and multiple second semantic recognition results corresponding to each feature marker in the feature marker screening set, and obtain a coral reef ecological recognition result of the area to be recognized according to the multiple first semantic recognition results and the multiple second semantic recognition results.

[0078] In this application, the remote sensing image is first divided according to multiple time period division methods, and then multiple time series feature markers are obtained by combining with a time series feature model, which can extract the changes of the ecological population of coral reefs within different time period ranges, considering both the ecological changes within a short time range and the ecological changes within a long time range, so as to improve the recognition accuracy and precision; then the remote sensing image is divided according to multiple grid division methods to obtain a first image set, a second image set and a third image set, and then multiple first markers, multiple second markers and multiple third markers are obtained by combining with a preset ecological feature dictionary for feature matching, which can identify the ecological features of coral reefs at different granularities and embed the ecological features into the markers, so as to achieve the purpose of identifying the ecological population of coral reefs; furthermore, multiple time series feature markers are corrected according to multiple first markers and multiple third markers, and a coral reef ecological recognition result to be recognized is obtained according to the multiple corrected feature markers and multiple second markers, which can combine the recognition results at different granularities and the recognition results with time series features, further reduce the recognition error, and thus improve the recognition accuracy and precision of the ecological population of coral reefs. Description of the Drawings

[0079] Figure 1 : is a schematic flowchart of a method for identifying the ecological diversity of underwater coral reefs shown in some embodiments of this application;

[0080] Figure 2 : is a module structure diagram of a device for identifying the ecological diversity of underwater coral reefs shown in some embodiments of this application. Detailed Embodiments

[0081] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, in which the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by combining the accompanying drawings are exemplary and are only used to explain some embodiments of the present application, and cannot be understood as a limitation on the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments shown in the present application without creative efforts belong to the protection scope of the present application.

[0082] In the description of the present application, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present application, unless otherwise specifically defined, the meaning of "a plurality" and "several" is two or more.

[0083] When identifying the ecological populations of coral reefs in the prior art, generally only the identification in a short-term range or instant data is considered, and the changes of ecological populations in a long-term range are not considered; at the same time, the identification of local detailed images is not considered, and both will lead to a decrease in the identification accuracy and precision. Therefore, how to improve the identification accuracy and precision of the ecological populations of coral reefs remains an urgent problem to be solved.

[0084] Based on the above technical background, please refer to Figure 1 , the embodiments of the present application provide an underwater coral reef ecological diversity identification method, including steps S101 to S105, and the specific steps are as follows:

[0085] Step S101: Obtain multiple remote sensing images of the area to be identified, and perform multiple divisions on the multiple remote sensing images according to a preset variety of time period division methods to obtain multiple sets of time series image sets; wherein, the variety of time period division methods include 90 days, 180 days, and 360 days.

[0086] In some embodiments of the present application, the obtaining of multiple remote sensing images of the area to be identified specifically includes:

[0087] Obtain multiple initial remote sensing images and multiple initial infrared images of the area to be identified;

[0088] Based on a preset feature recognition model, obtain the common area features of the multiple initial remote sensing images and the multiple initial infrared images;

[0089] According to the common area features, perform information feature fusion on the multiple initial remote sensing images and the multiple initial infrared images to obtain multiple remote sensing images.

[0090] This application first obtains an initial remote sensing image and an initial infrared image, obtains common area features through a feature recognition model, and performs information feature fusion on the initial remote sensing image and the initial infrared image based on the common area features to obtain a remote sensing image, which can improve the information density in the remote sensing image, and thus is conducive to subsequent feature recognition and feature extraction, and is more adaptable to the actual needs of the task.

[0091] In some embodiments of this application, the multiple remote sensing images are divided multiple times according to a preset variety of time period division methods to obtain multiple groups of time series image sets, specifically:

[0092] The multiple groups of time series image sets include a first time series image set, a second time series image set, and a third time series image set;

[0093] The multiple remote sensing images are divided into a first time series image set with a period of 90 days;

[0094] The multiple remote sensing images are divided into a second time series image set with a period of 180 days;

[0095] The multiple remote sensing images are divided into a third time series image set with a period of 360 days.

[0096] It should be understood that the variety of time period division methods can be set by those skilled in the art according to the actual situation for the number of types of time period division and the specific scope of time period division, and can also be set according to the experience of those skilled in the art for the number of types of time period division and the specific scope of time period division. It should be understood that the solutions of other time period division methods set by those skilled in the art without creative labor fall within the protection scope of this application.

[0097] Step S102: According to the multiple groups of time series image sets, in combination with a preset time series feature model, obtain multiple time series feature markers corresponding to the multiple remote sensing images.

[0098] In some embodiments of this application, the obtaining multiple time series feature markers corresponding to the multiple remote sensing images according to the multiple groups of time series image sets and in combination with a preset time series feature model specifically includes:

[0099] According to a preset training classification ratio, each of the multiple groups of time series image sets is divided into multiple groups of time series image training sets and corresponding multiple groups of time series image prediction sets;

[0100] According to the multiple groups of time series image training sets, perform iterative training on the time series feature model. In each round of iteration, input a group of time series image training sets corresponding to the current round into the time series feature model for iteration to adjust the model parameters until the number of iteration rounds reaches the number of groups of the multiple groups of time series image training sets, and obtain and output a first time series feature model;

[0101] Input the multiple sets of time-series image prediction sets into the first time-series feature model to obtain multiple time-series feature markers corresponding to the multiple remote sensing images.

[0102] In some embodiments of the present application, the preferred value of the preset training classification ratio is 3:1. That is, in each set of time-series image sets, 75% of the time-series image data is the time-series image training set, and 25% of the time-series image data is the time-series image prediction set.

[0103] In some embodiments of the present application, the time-series feature model includes, but is not limited to, a long short-term memory network (LSTM) model, a bidirectional recurrent neural network (Bi-RNN) model, a deep recurrent neural network (Deep RNN) model, and a temporal convolutional network (TCN) model.

[0104] In the present application, multiple sets of time-series image sets are first divided into a training set and a prediction set according to a preset training classification ratio, and then the time-series feature model is iteratively trained multiple times based on the training set, and multiple time-series feature markers corresponding to multiple remote sensing images are obtained based on the prediction set. This can enable the time-series feature model to gradually learn the characteristics of the changes in the ecological population of coral reefs, so as to extract the changes in the ecological population of coral reefs within different time periods, and improve the recognition accuracy and precision.

[0105] Step S103: Perform multiple divisions on the multiple remote sensing images according to a preset variety of grid division methods to obtain a first image set, a second image set, and a third image set; wherein, the variety of grid division methods are arranged in decreasing order of grid size and are divided into a first-class grid, a second-class grid, and a third-class grid; the first image set is divided according to the first-class grid, the second image set is divided according to the second-class grid, and the third image set is divided according to the third-class grid.

[0106] In some embodiments of the present application, the performing multiple divisions on the multiple remote sensing images according to a preset variety of grid division methods to obtain a first image set, a second image set, and a third image set specifically includes:

[0107] Divide the multiple remote sensing images according to the grid size of the first-class grid to obtain a first grid image set;

[0108] Divide the multiple remote sensing images according to the grid size of the second-class grid to obtain a second grid image set;

[0109] Divide the multiple remote sensing images according to the grid sizes of the three types of grids to obtain an initial third grid image set, and perform feature enhancement on the initial third grid image set to obtain a third grid image set;

[0110] Perform denoising processing on the first grid image set, the second grid image set, and the third grid image set, and align the image feature information to obtain a first image set, a second image set, and a third image set.

[0111] In some embodiments of the present application, the preferred solutions for the first type of grid, the second type of grid, and the third type of grid are . Specifically, taking the preferred solution of the first type of grid as an example, That is, the current remote sensing image is cut into a grid array with 1280 columns horizontally and 720 rows vertically, and the images of each grid in the grid array are integrated to obtain a first grid image subset corresponding to the current remote sensing image. Specifically, the union of multiple first grid image subsets obtained by processing the multiple remote sensing images is used as the first grid image set.

[0112] In the present application, the multiple remote sensing images are first divided according to the grid sizes of the first type of grid and the second type of grid to obtain a first grid image set and a second grid image set, and then the multiple remote sensing images are divided according to the grid size of the third type of grid and feature enhanced to obtain a third grid image set, which can prevent the third grid image set divided by the grid size of the third type of grid from losing picture feature details. Furthermore, denoising and aligning the image feature information for the first grid image set, the second grid image set, and the third grid image set can divide the remote sensing images at different granularities, thereby meeting the subsequent requirements for distinguishing coral reef ecological population information at different granularities and being more adaptable to the actual task requirements.

[0113] Step S104: According to the first image set, the second image set, and the third image set, perform feature matching in combination with a preset ecological feature dictionary to obtain multiple first markers, multiple second markers, and multiple third markers.

[0114] In some embodiments of the present application, the performing feature matching according to the first image set, the second image set, and the third image set in combination with a preset ecological feature dictionary to obtain multiple first markers, multiple second markers, and multiple third markers specifically includes:

[0115] According to the first image set, the second image set, and the third image set, in combination with a preset feature extraction coding method, obtain multiple first semantic features, multiple second semantic features, and multiple third semantic features;

[0116] Based on the similarity algorithm, perform feature matching on the multiple first semantic features, the multiple second semantic features, and the multiple third semantic features according to the preset ecological feature dictionary to obtain multiple first tags, multiple second tags, and multiple third tags.

[0117] In some embodiments of the present application, the feature extraction and encoding method specifically includes:

[0118] Extract features from each image in the current image set to obtain multiple image feature vectors;

[0119] Perform semantic recognition encoding on the multiple image feature vectors to obtain multiple semantic features.

[0120] In some embodiments of the present application, the preset ecological feature dictionary includes the categories of ecological populations of coral reefs and corresponding features.

[0121] In some embodiments of the present application, the preferred scheme of the model based on which the semantic recognition encoding is performed is the LSeg model or the CAT-Seg model.

[0122] In some embodiments of the present application, the similarity algorithm includes but is not limited to the cosine similarity algorithm, the Pearson correlation coefficient algorithm, and the Jaccard similarity coefficient algorithm.

[0123] The present application first extracts multiple first semantic features, multiple second semantic features, and multiple third semantic features through the feature extraction and encoding method, and then performs feature matching based on the similarity algorithm according to the preset ecological feature dictionary to obtain multiple first tags, multiple second tags, and multiple third tags, which can identify the ecological features of coral reefs at different granularities and embed the ecological features into the tags, achieving the purpose of identifying the ecological populations of coral reefs.

[0124] Step S105: According to the multiple first tags and the multiple third tags, correct the multiple timing feature tags to obtain multiple corrected feature tags, and according to the multiple corrected feature tags and the multiple second tags, obtain the coral reef ecological recognition result of the area to be recognized.

[0125] In some embodiments of the present application, the correcting the multiple timing feature tags to obtain multiple corrected feature tags according to the multiple first tags and the multiple third tags specifically includes:

[0126] Eliminate the timing feature tags in the multiple timing feature tags that do not belong to the areas corresponding to the multiple first tags to obtain multiple initial corrected feature tags;

[0127] Stitch the multiple third tags according to the grid size of the one type of grid corresponding to the multiple first tags to obtain multiple grid areas;

[0128] Eliminate the initial correction feature markers that do not belong to the multiple grid regions among the multiple initial correction feature markers to obtain multiple correction feature markers.

[0129] In this application, first, eliminate the time-series feature markers that do not belong to the regions corresponding to the multiple first markers among the multiple time-series feature markers to obtain multiple initial correction feature markers, which can screen the time-series feature markers at the corresponding granularity of the first markers, and filter out some misrecognized time-series feature markers. Then, splice multiple third markers according to the grid size of a type of grid corresponding to the first markers to obtain multiple grid regions, which can simplify the process of screening and eliminating through the third markers and improve the screening efficiency. Then, eliminate the initial correction feature markers that do not belong to the multiple grid regions among the multiple initial correction feature markers to obtain multiple correction feature markers, which can screen the time-series feature markers at the corresponding granularity of the third markers and further filter out the misrecognized items in the time-series feature markers, thereby improving the recognition accuracy and precision of the ecological population of coral reefs.

[0130] In some embodiments of the present application, obtaining the coral reef ecological recognition result of the region to be recognized according to the multiple correction feature markers and the multiple second markers specifically includes:

[0131] Obtain a feature marker intersection and a feature marker symmetric difference set according to the multiple correction feature markers and the multiple second markers;

[0132] Screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold to obtain a feature marker screening set;

[0133] Obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection and multiple second semantic recognition results corresponding to each feature marker in the feature marker screening set, and obtain the coral reef ecological recognition result of the region to be recognized according to the multiple first semantic recognition results and the multiple second semantic recognition results.

[0134] In this application, first obtain a feature marker intersection and a feature marker symmetric difference set according to multiple correction feature markers and multiple second markers, then filter out the feature markers in the feature marker symmetric difference set whose recognition probabilities are not greater than a preset recognition threshold to obtain a feature marker screening set, which can further exclude misrecognized feature markers. Then, obtain the coral reef ecological recognition result of the region to be recognized according to the semantic recognition results of the feature marker intersection and the feature marker screening set, which can integrate the recognition results of different granularities and the recognition results with time-series features, and further improve the recognition accuracy and precision of the ecological population of coral reefs.

[0135] Compared with the prior art, in this application, the remote sensing images are first divided according to multiple time period division methods, and then multiple time series feature markers are obtained by combining with the time series feature model, which can extract the changes of the ecological population of coral reefs within different time periods, taking into account both the ecological changes in the short term and the long term, thereby improving the recognition accuracy and precision. Then, the remote sensing images are divided according to multiple grid division methods to obtain the first image set, the second image set, and the third image set, and then multiple first markers, multiple second markers, and multiple third markers are obtained by combining with a preset ecological feature dictionary, which can identify the ecological features of coral reefs at different granularities and embed the ecological features into the markers to achieve the purpose of identifying the ecological population of coral reefs. Furthermore, multiple time series feature markers are corrected according to multiple first markers and multiple third markers, and the ecological recognition result of the coral reef to be identified is obtained according to multiple corrected feature markers and multiple second markers, which can combine the recognition results at different granularities and the recognition results with time series features, further reducing the recognition error, thereby improving the recognition accuracy and precision of the ecological population of coral reefs.

[0136] Corresponding to the foregoing method, please refer to Figure 2 , an underwater coral reef ecological diversity recognition system is provided in an embodiment of this application, including a time series division module 210, a time series marking module 220, a grid division module 230, a feature matching module 240, and a result obtaining module 250;

[0137] The time series division module 210 is configured to obtain multiple remote sensing images of the area to be recognized, and perform multiple divisions on the multiple remote sensing images according to a preset multiple time period division methods to obtain multiple groups of time series image sets; wherein, the multiple time period division methods include 90 days, 180 days, and 360 days;

[0138] The time series marking module 220 is configured to obtain multiple time series feature markers corresponding to the multiple remote sensing images by combining with a preset time series feature model according to the multiple groups of time series image sets;

[0139] The grid division module 230 is configured to perform multiple divisions on the multiple remote sensing images according to a preset multiple grid division methods to obtain a first image set, a second image set, and a third image set; wherein, the multiple grid division methods are arranged in decreasing order of grid size and are divided into a first-class grid, a second-class grid, and a third-class grid; the first image set is divided according to the first-class grid, the second image set is divided according to the second-class grid, and the third image set is divided according to the third-class grid;

[0140] The feature matching module 240 is configured to perform feature matching by combining with a preset ecological feature dictionary according to the first image set, the second image set, and the third image set to obtain multiple first markers, multiple second markers, and multiple third markers;

[0141] The result acquisition module 250 is configured to correct the timing feature markers according to the multiple first markers and the multiple third markers to obtain multiple corrected feature markers, and obtain the coral reef ecological recognition result of the area to be recognized according to the multiple corrected feature markers and the multiple second markers.

[0142] In some embodiments of the present application, the timing division module 210 includes an image acquisition sub-module 211; the image acquisition sub-module includes an image acquisition unit 2111, a feature recognition unit 2112, and a feature fusion unit 2113;

[0143] The image acquisition unit 2111 is configured to acquire multiple initial remote sensing images and multiple initial infrared images of the area to be recognized;

[0144] The feature recognition unit 2112 is configured to obtain the common area features of the multiple initial remote sensing images and the multiple initial infrared images based on a preset feature recognition model;

[0145] The feature fusion unit 2113 is configured to perform information feature fusion on the multiple initial remote sensing images and the multiple initial infrared images according to the common area features to obtain multiple remote sensing images.

[0146] In some embodiments of the present application, the timing marking module 220 includes a data division sub-module 221, a model training sub-module 222, and a timing marking sub-module 223;

[0147] The data division sub-module 221 is configured to divide each of the multiple groups of timing image sets into multiple groups of timing image training sets and corresponding multiple groups of timing image prediction sets according to a preset training classification ratio;

[0148] The model training sub-module 222 is configured to perform iterative training on the timing feature model according to the multiple groups of timing image training sets. In each iteration, a group of timing image training sets corresponding to the current iteration is input into the timing feature model for iteration to adjust the model parameters until the iteration round reaches the number of groups of the multiple groups of timing image training sets, and obtain and output a first timing feature model;

[0149] The timing marking sub-module 223 is configured to input the multiple groups of timing image prediction sets into the first timing feature model to obtain multiple timing feature markers corresponding to the multiple remote sensing images.

[0150] In some embodiments of the present application, the grid division module 230 includes a first division sub-module 231, a second division sub-module 232, a third division sub-module 233, and a feature processing sub-module 234;

[0151] The first sub - division module 231 is configured to divide the multiple remote - sensing images according to the grid size of the first - type grid to obtain a first grid image set;

[0152] The second sub - division module 232 is configured to divide the multiple remote - sensing images according to the grid size of the second - type grid to obtain a second grid image set;

[0153] The third sub - division module 233 is configured to divide the multiple remote - sensing images according to the grid size of the third - type grid to obtain an initial third grid image set, and perform feature enhancement on the initial third grid image set to obtain a third grid image set;

[0154] The feature processing sub - module 234 is configured to perform denoising processing on the first grid image set, the second grid image set, and the third grid image set, and align the image feature information to obtain a first image set, a second image set, and a third image set.

[0155] In some embodiments of the present application, the feature matching module 240 includes a feature extraction sub - module 241 and a feature matching sub - module 242;

[0156] The feature extraction sub - module 241 is configured to obtain a plurality of first semantic features, a plurality of second semantic features, and a plurality of third semantic features according to the first image set, the second image set, and the third image set, in combination with a preset feature extraction coding method;

[0157] The feature matching sub - module 242 is configured to perform feature matching on the plurality of first semantic features, the plurality of second semantic features, and the plurality of third semantic features based on a similarity algorithm according to a preset ecological feature dictionary to obtain a plurality of first tags, a plurality of second tags, and a plurality of third tags.

[0158] In some embodiments of the present application, the feature extraction coding method specifically includes:

[0159] Perform feature extraction on each image of the current image set to obtain a plurality of image feature vectors;

[0160] Perform semantic recognition coding on the plurality of image feature vectors to obtain a plurality of semantic features.

[0161] In some embodiments of the present application, the result acquisition module 250 includes a feature correction sub - module 251; the feature correction sub - module 251 includes an initial correction unit 2511, a grid splicing unit 2512, and a feature correction unit 2513;

[0162] The initial correction unit 2511 is configured to eliminate the timing feature markers that do not belong to the regions corresponding to the multiple first markers among the multiple timing feature markers, so as to obtain multiple initial correction feature markers;

[0163] The grid splicing unit 2512 is configured to splice the multiple third markers according to the grid size of the class of grids corresponding to the multiple first markers, so as to obtain multiple grid regions;

[0164] The feature correction unit 2513 is configured to eliminate the initial correction feature markers that do not belong to the multiple grid regions among the multiple initial correction feature markers, so as to obtain multiple correction feature markers.

[0165] In some embodiments of the present application, the result acquisition module 250 includes a result acquisition sub-module 252; the result acquisition sub-module 252 includes a marker aggregation unit 2521, a marker screening unit 2522, and a result acquisition unit 2523;

[0166] The marker aggregation unit 2521 is configured to obtain a feature marker intersection and a feature marker symmetric difference set according to the multiple correction feature markers and the multiple second markers;

[0167] The marker screening unit 2522 is configured to screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold, so as to obtain a feature marker screening set;

[0168] The result acquisition unit 2523 is configured to obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection and multiple second semantic recognition results corresponding to each feature marker in the feature marker screening set, and obtain a coral reef ecological recognition result of the region to be recognized according to the multiple first semantic recognition results and the multiple second semantic recognition results.

[0169] This application first divides the remote sensing image according to multiple time period division methods, and then combines the time series feature model to obtain multiple time series feature markers, which can extract the changes of the ecological population of coral reefs within different time periods, taking into account both the ecological changes within a short time range and those within a long time range, thereby improving the recognition accuracy; then divides the remote sensing image according to multiple grid division methods to obtain the first image set, the second image set, and the third image set, and then combines the preset ecological feature dictionary for feature matching to obtain multiple first markers, multiple second markers, and multiple third markers, which can identify the ecological features of coral reefs at different granularities and embed the ecological features into the markers to achieve the purpose of identifying the ecological population of coral reefs; furthermore, corrects multiple time series feature markers according to multiple first markers and multiple third markers, and obtains the ecological recognition result of the coral reef to be recognized according to multiple corrected feature markers and multiple second markers, which can combine the recognition results of different granularities and the recognition results with time series features to further reduce the recognition error, thereby improving the recognition accuracy and accuracy of the ecological population of coral reefs.

[0170] Adaptively, the embodiment of this application also provides a computer device and a computer-readable storage medium.

[0171] The computer device includes: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor;

[0172] Wherein, when the processor executes the computer program, it implements an underwater coral reef ecological diversity recognition method of this application.

[0173] The computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by the processor to execute an underwater coral reef ecological diversity recognition method of this application.

[0174] The above is a partial embodiment of this application, which further details the purpose, technical solution, and beneficial effects of this application. It should be clear that the above partial embodiment of this application should not be construed as a limitation of this application. In particular, for those skilled in the art, any changes, modifications, equivalent replacements, and variations made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. An underwater coral reef ecological diversity identification method, characterized in that, Including: Obtain multiple remote sensing images of the area to be recognized, and perform multiple divisions on the multiple remote sensing images according to a preset variety of time period division methods to obtain multiple groups of time series image sets; wherein, the variety of time period division methods include 90 days, 180 days, and 360 days; According to the multiple groups of time series image sets, in combination with a preset time series feature model, obtain multiple time series feature marks corresponding to the multiple remote sensing images; Perform multiple divisions on the multiple remote sensing images according to a preset variety of grid division methods to obtain a first image set, a second image set, and a third image set; wherein, the variety of grid division methods are arranged in descending order of grid size and are divided into first-class grids, second-class grids, and third-class grids; the first image set is divided according to the first-class grids, the second image set is divided according to the second-class grids, and the third image set is divided according to the third-class grids; According to the first image set, the second image set, and the third image set, perform feature matching in combination with a preset ecological feature dictionary to obtain multiple first marks, multiple second marks, and multiple third marks; According to the multiple first marks and the multiple third marks, correct the multiple time series feature marks to obtain multiple corrected feature marks, and according to the multiple corrected feature marks and the multiple second marks, obtain the coral reef ecological recognition result of the area to be recognized; The performing feature matching according to the first image set, the second image set, and the third image set in combination with a preset ecological feature dictionary to obtain multiple first marks, multiple second marks, and multiple third marks specifically includes: According to the first image set, the second image set, and the third image set, in combination with a preset feature extraction coding method, obtain multiple first semantic features, multiple second semantic features, and multiple third semantic features; Based on a similarity algorithm, perform feature matching on the multiple first semantic features, the multiple second semantic features, and the multiple third semantic features according to a preset ecological feature dictionary to obtain multiple first marks, multiple second marks, and multiple third marks; The obtaining the coral reef ecological recognition result of the area to be recognized according to the multiple corrected feature marks and the multiple second marks specifically includes: According to the multiple corrected feature marks and the multiple second marks, obtain a feature mark intersection and a feature mark symmetric difference set; Screen the feature marks in the feature mark symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold to obtain a feature mark screening set; Obtain multiple first semantic recognition results corresponding to each feature mark in the feature mark intersection and multiple second semantic recognition results corresponding to each feature mark in the feature mark screening set, and according to the multiple first semantic recognition results and the multiple second semantic recognition results, obtain the coral reef ecological recognition result of the area to be recognized.

2. The method for identifying the ecological diversification of underwater coral reefs according to claim 1, characterized in that, The obtaining multiple remote sensing images of the area to be recognized specifically includes: Obtain multiple initial remote sensing images and multiple initial infrared images of the area to be recognized; Based on a preset feature recognition model, obtain the common area features of the multiple initial remote sensing images and the multiple initial infrared images; According to the common area features, perform information feature fusion on the multiple initial remote sensing images and the multiple initial infrared images to obtain multiple remote sensing images.

3. The method for identifying the ecological diversification of underwater coral reefs according to claim 1, characterized in that The multiple remote sensing images are divided multiple times according to a variety of preset grid division methods to obtain a first image set, a second image set, and a third image set. Specifically, it includes: Divide the multiple remote sensing images according to the grid size of the first type of grid to obtain a first grid image set; Divide the multiple remote sensing images according to the grid size of the second type of grid to obtain a second grid image set; Divide the multiple remote sensing images according to the grid size of the third type of grid to obtain an initial third grid image set, and perform feature enhancement on the initial third grid image set to obtain a third grid image set; Perform denoising processing on the first grid image set, the second grid image set, and the third grid image set, and align the image feature information to obtain a first image set, a second image set, and a third image set.

4. The method for identifying the ecological diversity of underwater coral reefs according to claim 1, characterized in that, The feature extraction and encoding method specifically includes: Extract features from each image in the current image set to obtain multiple image feature vectors; Perform semantic recognition encoding on the multiple image feature vectors to obtain multiple semantic features.

5. The method for identifying the ecological diversity of underwater coral reefs according to claim 1, characterized in that The coral reef ecological recognition result of the area to be recognized is obtained according to the multiple corrected feature markers and the multiple second markers. Specifically, it includes: According to the multiple corrected feature markers and the multiple second markers, obtain a feature marker intersection and a feature marker symmetric difference set; Screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than the preset recognition threshold to obtain a feature marker screening set; Obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection and multiple second semantic recognition results corresponding to each feature marker in the feature marker screening set, and obtain the coral reef ecological recognition result of the area to be recognized according to the multiple first semantic recognition results and the multiple second semantic recognition results.

6. An underwater coral reef ecological diversity recognition system, characterized in that, It includes a time series division module, a time series marking module, a grid division module, a feature matching module, and a result acquisition module; The time series division module is used to obtain multiple remote sensing images of the area to be recognized, and divide the multiple remote sensing images multiple times according to a variety of preset time period division methods to obtain multiple groups of time series image sets; among them, the variety of time period division methods include 90 days, 180 days, and 360 days; The time series marking module is used to obtain multiple time series feature markers corresponding to the multiple remote sensing images according to the multiple groups of time series image sets in combination with a preset time series feature model; The grid division module is used to divide the multiple remote sensing images multiple times according to a variety of preset grid division methods to obtain a first image set, a second image set, and a third image set; among them, the variety of grid division methods are arranged in descending order of grid size and divided into a first type of grid, a second type of grid, and a third type of grid; the first image set is divided according to the first type of grid, the second image set is divided according to the second type of grid, and the third image set is divided according to the third type of grid; The feature matching module is configured to perform feature matching based on the first image set, the second image set, and the third image set, in combination with a preset ecological feature dictionary, to obtain a plurality of first markers, a plurality of second markers, and a plurality of third markers; The result obtaining module is configured to correct the timing feature markers according to the plurality of first markers and the plurality of third markers to obtain a plurality of corrected feature markers, and obtain a coral reef ecological recognition result of the area to be recognized according to the plurality of corrected feature markers and the plurality of second markers; The result obtaining module includes a result obtaining sub-module; the result obtaining sub-module includes a marker aggregation unit, a marker screening unit, and a result obtaining unit; The marker aggregation unit is configured to obtain a feature marker intersection and a feature marker symmetric difference set according to the plurality of corrected feature markers and the plurality of second markers; The marker screening unit is configured to screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold to obtain a feature marker screening set; The result obtaining unit is configured to obtain a plurality of first semantic recognition results corresponding to each feature marker in the feature marker intersection and a plurality of second semantic recognition results corresponding to each feature marker in the feature marker screening set, and obtain a coral reef ecological recognition result of the area to be recognized according to the plurality of first semantic recognition results and the plurality of second semantic recognition results; The step of performing feature matching based on the first image set, the second image set, and the third image set, in combination with a preset ecological feature dictionary, to obtain a plurality of first markers, a plurality of second markers, and a plurality of third markers specifically includes: Obtaining a plurality of first semantic features, a plurality of second semantic features, and a plurality of third semantic features according to the first image set, the second image set, and the third image set, in combination with a preset feature extraction and encoding method; Performing feature matching on the plurality of first semantic features, the plurality of second semantic features, and the plurality of third semantic features based on a similarity algorithm according to a preset ecological feature dictionary to obtain a plurality of first markers, a plurality of second markers, and a plurality of third markers.

Images