Underwater coral reef ecological diversification identification method and system

Abstract

The invention discloses an underwater coral reef ecological diversification identification method and system, and relates to the field of marine ecological analysis, and the method comprises the steps: obtaining a plurality of remote sensing images of a to-be-identified region, and carrying out the multi-time division of the remote sensing images according to a plurality of preset time division modes, and obtaining a plurality of time sequence image sets; according to the multiple groups of time sequence image sets, obtaining a plurality of time sequence feature marks in combination with a time sequence feature model; dividing the plurality of remote sensing images for multiple times according to multiple grid division modes 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, performing feature matching in combination with a preset ecological feature dictionary to obtain a plurality of first marks, a plurality of second marks and a plurality of third marks; and according to the plurality of first marks and the plurality of third marks, correcting the plurality of time sequence feature marks to obtain a plurality of corrected feature marks, and according to the plurality of corrected feature marks and the plurality of second marks, obtaining a coral reef ecological recognition result of the to-be-recognized area.

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: 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; 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; 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; According to the first image set, the second image set and the third image set, 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; According to the multiple first markers and the multiple third markers, the multiple time series feature markers are corrected to obtain multiple corrected feature markers, and according to the multiple corrected feature markers and the multiple second markers, a coral reef ecological identification result of the area to be identified is obtained.

[0006] The present application first divides the remote sensing image according to a plurality of time period division methods, and then obtains a plurality of time series feature markers in combination with a time series feature model, which can extract the changes of ecological populations of coral reefs within different time periods, taking into account both the ecological changes within a short time range and the ecological changes within a long time range, thereby improving the recognition precision and accuracy; then the remote sensing image is divided according to a plurality of grid division methods to obtain a first image set, a second image set and a third image set, and then 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 characteristics of coral reefs at different granularities and embed the ecological characteristics into the markers, thereby achieving the purpose of identifying the ecological populations of coral reefs; then, the plurality of time series feature markers are corrected according to the plurality of first markers and the plurality of third markers, and the ecological recognition results of the coral reefs to be identified are obtained according to the plurality of corrected feature markers and the plurality of second markers, which can combine the recognition results of different granularities and the recognition results with time series features, further reduce the recognition error, thereby improving the recognition precision and accuracy of the ecological populations of coral reefs.

[0007] In certain embodiments of the present application, the acquiring of multiple remote sensing images of the area to be identified specifically includes: Acquire multiple initial remote sensing images and multiple initial infrared images of the area to be identified; Based on a preset feature recognition model, common regional features of the multiple initial remote sensing images and the multiple initial infrared images are obtained; According to the common area features, the multiple initial remote sensing images and the multiple initial infrared images are fused with information features to obtain multiple remote sensing images.

[0008] This application first obtains the initial remote sensing image and the initial infrared image, obtains the common area features through the feature recognition model, and obtains the remote sensing image by fusing the information features of the initial remote sensing image and the initial infrared image based on the common area features. This can improve the information density in the remote sensing image, which is beneficial to the subsequent feature recognition and feature extraction, and is more suitable for the actual needs of the task.

[0009] In certain embodiments of the present application, the multiple time series feature labels corresponding to the multiple remote sensing images are obtained based on the multiple groups of time series image sets in combination with a preset time series feature model, specifically including: 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; Iteratively train the time series feature model according to the multiple groups of time series image training sets, input a group of time series image training sets of the corresponding round into the time series feature model for iteration in each round of iteration to adjust 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; The multiple groups of time series image prediction sets are input into the first time series feature model to obtain multiple time series feature labels corresponding to the multiple remote sensing images.

[0010] The present application first divides multiple groups of time series image sets into training sets and prediction sets according to a preset training classification ratio, then performs multiple iterative training on the time series feature model based on the training set, and obtains multiple time series feature labels corresponding to multiple remote sensing images based on the prediction set, which enables the time series feature model to gradually learn the characteristics of changes in the ecological population of coral reefs, thereby being able to extract changes in the ecological population of coral reefs within different time periods and improve recognition precision and accuracy.

[0011] In certain embodiments of the present application, the multiple remote sensing images are divided multiple times according to the preset multiple grid division methods to obtain the first image set, the second image set and the third image set, specifically including: Dividing the plurality of remote sensing images according to the grid size of the first type of grid to obtain a first grid image set; Dividing the plurality of remote sensing images according to the grid sizes of the second type of grids to obtain a second grid image set; Dividing the plurality of remote sensing images according to the grid sizes of the three types of grids 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; The first grid image set, the second grid image set and the third grid image set are subjected to denoising processing, and image feature information is aligned to obtain a first image set, a second image set and a third image set.

[0012] The present application first divides multiple remote sensing images according to the grid sizes of the first and second grids to obtain a first grid image set and a second grid image set, and then divides the multiple remote sensing images according to the grid sizes of the third grid and performs feature enhancement to obtain a third grid image set. This can prevent the third grid image set divided by the grid sizes of the three grids from losing picture feature details, and then denoises the first grid image set, the second grid image set, and the third grid image set and aligns image feature information. This can divide remote sensing images at different granularities, thereby meeting the subsequent need to distinguish coral reef ecological population information at different granularities, and is more adapted to the actual needs of the task.

[0013] In certain embodiments of the present application, the feature matching is performed 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 including: 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, a plurality of first semantic features, a plurality of second semantic features and a plurality of third semantic features are obtained; 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 tags, multiple second tags, and multiple third tags.

[0014] In certain embodiments of the present application, the feature extraction encoding method specifically includes: Extract features from each image in the current image set to obtain multiple image feature vectors; Semantic recognition coding is performed on the multiple image feature vectors to obtain multiple semantic features.

[0015] The present application first extracts multiple first semantic features, multiple second semantic features and multiple third semantic features through a feature extraction and coding method, and then based on a similarity algorithm, performs feature matching according to a preset ecological feature dictionary to obtain multiple first markers, multiple second markers and multiple third markers. The application can identify the ecological characteristics of coral reefs at different granularities and embed the ecological characteristics into the markers, thereby achieving the purpose of identifying coral reef ecological populations.

[0016] In certain embodiments of the present application, the correcting the plurality of timing feature marks to obtain a plurality of corrected feature marks according to the plurality of first marks and the plurality of third marks specifically includes: Eliminate the timing feature marks that do not belong to the regions corresponding to the multiple first marks from the multiple timing feature marks to obtain multiple initial corrected feature marks; According to the grid size of the one type of grid corresponding to the multiple first marks, splicing the multiple third marks to obtain multiple grid areas; Initial revised feature marks that do not belong to the multiple grid areas are eliminated from the multiple initial revised feature marks to obtain multiple revised feature marks.

[0017] The present application first eliminates the time series feature markers that do not belong to the corresponding areas of the multiple first markers from the multiple time series feature markers to obtain multiple initial corrected feature markers, can screen the time series feature markers at the corresponding granularity of the first marker, screen out a part of the time series feature markers with identification errors, and then splice the multiple third markers according to the grid size of a type of grid corresponding to the first marker to obtain multiple grid areas, which can simplify the process of screening and eliminating through the third marker and improve the screening efficiency, and then eliminate the initial corrected feature markers that do not belong to the multiple grid areas from the multiple initial corrected feature markers to obtain multiple corrected feature markers, can screen the time series feature markers at the corresponding granularity of the third marker, and further screen out the identification errors in the time series feature markers, thereby improving the recognition accuracy and precision of the ecological populations of coral reefs.

[0018] In certain embodiments of the present application, obtaining the coral reef ecological identification result of the area to be identified based on the multiple corrected characteristic markers and the multiple second markers specifically includes: Obtaining a feature marker intersection set and a feature marker symmetric difference set according to the multiple modified feature markers and the multiple second markers; Selecting feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold value to obtain a feature marker screening set; A plurality of first semantic recognition results corresponding to each feature marker of the feature marker intersection set and a plurality of second semantic recognition results corresponding to each feature marker of the feature marker screening set are obtained, and a coral reef ecological recognition result of the area to be identified is obtained based on the plurality of first semantic recognition results and the plurality of second semantic recognition results.

[0019] The present application first obtains a feature marker intersection and a feature marker symmetric difference set based on multiple corrected feature markers and multiple second markers, and then screens out feature markers in the feature marker symmetric difference set whose identification probability is not greater than a preset identification threshold to obtain a feature marker screening set, which can further eliminate feature markers with identification errors, and then obtains the coral reef ecological identification result of the area to be identified based on the semantic recognition results of the feature marker intersection and the feature marker screening set, which can combine the identification results of different granularities and the identification results with time series characteristics to further improve the identification precision and accuracy of the ecological populations of coral reefs.

[0020] According to a second aspect of the implementation of the present application, there is provided an underwater coral reef ecological diversity identification system, including 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 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; The time series marking module is used to obtain a plurality of time series feature marks corresponding to the plurality of remote sensing images according to the plurality of 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 multiple preset 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 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; The feature matching module is used 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 acquisition module is used to correct the time series feature marks to obtain multiple corrected feature marks according to the multiple first marks and the multiple third marks, and obtain the coral reef ecological identification result of the area to be identified according to the multiple corrected feature marks and the multiple second marks.

[0021] In certain embodiments of the present application, the time division module includes an image acquisition submodule; the image acquisition submodule includes an image acquisition unit, a feature recognition unit and a feature fusion unit; The image acquisition unit is used to acquire a plurality of initial remote sensing images and a plurality of initial infrared images of the area to be identified; The feature recognition unit is used to obtain common area features of the multiple initial remote sensing images and the multiple initial infrared images based on a preset feature recognition model; The feature fusion unit is used to fuse the information features of the multiple initial remote sensing images and the multiple initial infrared images according to the common area features to obtain multiple remote sensing images.

[0022] In certain embodiments of the present application, the timing marking module includes a data partitioning submodule, a model training submodule and a timing marking submodule; The data division submodule is used to divide the multiple groups of time series image sets into multiple groups of time series image training sets and corresponding multiple groups of time series image prediction sets according to a preset training classification ratio; The model training submodule is used to iteratively train the time series feature model according to the multiple groups of time series image training sets, and in each round of iteration, a group of time series image training sets of the corresponding round is input 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; The time series marking submodule is used to input the multiple groups of time series image prediction sets into the first time series feature model to obtain multiple time series feature marks corresponding to the multiple remote sensing images.

[0023] In certain embodiments of the present application, the grid partitioning module includes a first partitioning submodule, a second partitioning submodule, a third partitioning submodule, and a feature processing submodule; The first division submodule is used to divide the plurality of remote sensing images according to the grid size of the first type of grid to obtain a first grid image set; The second division submodule is used to divide the plurality of remote sensing images according to the grid sizes of the two types of grids to obtain a second grid image set; The third division submodule is used to 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; 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.

[0024] In certain embodiments of the present application, the feature matching module includes a feature extraction submodule and a feature matching submodule; 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; 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.

[0025] In certain embodiments of the present application, the feature extraction encoding method specifically includes: Extract features from each image in the current image set to obtain multiple image feature vectors; Semantic recognition coding is performed on the multiple image feature vectors to obtain multiple semantic features.

[0026] 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; 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; 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; 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.

[0027] 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; The tag aggregation unit is used to obtain a feature tag intersection set and a feature tag symmetric difference set according to the multiple modified feature tags and the multiple second tags; The marker screening unit is used to screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold value, to obtain a feature marker screening set; The result acquisition unit is used to obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection set 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 identified based on the multiple first semantic recognition results and the multiple second semantic recognition results.

[0028] The present application first divides the remote sensing image according to a plurality of time period division methods, and then obtains a plurality of time series feature markers in combination with a time series feature model, which can extract the changes of ecological populations of coral reefs within different time periods, taking into account both the ecological changes within a short time range and the ecological changes within a long time range, thereby improving the recognition precision and accuracy; then the remote sensing image is divided according to a plurality of grid division methods to obtain a first image set, a second image set and a third image set, and then 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 characteristics of coral reefs at different granularities and embed the ecological characteristics into the markers, thereby achieving the purpose of identifying the ecological populations of coral reefs; then, the plurality of time series feature markers are corrected according to the plurality of first markers and the plurality of third markers, and the ecological recognition results of the coral reefs to be identified are obtained according to the plurality of corrected feature markers and the plurality of second markers, which can combine the recognition results of different granularities and the recognition results with time series features, further reduce the recognition error, thereby improving the recognition precision and accuracy of the ecological populations of coral reefs. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1: A schematic diagram of a process for identifying underwater coral reef ecological diversity according to certain embodiments of the present application; Figure 2 : A module structure diagram of an underwater coral reef ecological diversity identification device shown in certain embodiments of the present application. DETAILED DESCRIPTION

[0030] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below in conjunction with the accompanying drawings are exemplary and are only used to explain some embodiments of the present application, and should not be construed as limitations on the embodiments of the present application. Based on the embodiments shown in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0031] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, unless otherwise clearly and specifically defined, "multiple" and "several" mean two or more.

[0032] When identifying ecological populations of coral reefs in the prior art, only short-term or instantaneous data is generally considered, and the changes of ecological populations in the long term are not considered; at the same time, the recognition of local detail images is not considered, both of which will lead to a decrease in recognition precision and accuracy. Therefore, how to improve the recognition precision and accuracy of ecological populations of coral reefs is still an urgent problem to be solved.

[0033] Based on the above technical background, please refer to Figure 1 The embodiment of the present application provides a method for identifying underwater coral reef ecological diversity, including steps S101 to S105, each of which is specifically as follows: Step S101: 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.

[0034] In certain embodiments of the present application, the acquiring of multiple remote sensing images of the area to be identified specifically includes: Acquire multiple initial remote sensing images and multiple initial infrared images of the area to be identified; Based on a preset feature recognition model, common regional features of the multiple initial remote sensing images and the multiple initial infrared images are obtained; According to the common area features, the multiple initial remote sensing images and the multiple initial infrared images are fused with information features to obtain multiple remote sensing images.

[0035] This application first obtains the initial remote sensing image and the initial infrared image, obtains the common area features through the feature recognition model, and obtains the remote sensing image by fusing the information features of the initial remote sensing image and the initial infrared image based on the common area features. This can improve the information density in the remote sensing image, which is beneficial to the subsequent feature recognition and feature extraction, and is more suitable for the actual needs of the task.

[0036] In certain embodiments of the present application, the plurality of remote sensing images are divided multiple times according to the preset multiple time period division methods to obtain multiple groups of time series image sets, specifically: The multiple groups of time-sequential image sets include a first time-sequential image set, a second time-sequential image set and a third time-sequential image set; Dividing the plurality of remote sensing images into a first time series image set with a period of 90 days; Dividing the plurality of remote sensing images into a second time series image set with a period of 180 days; The plurality of remote sensing images are divided into a third time series image set based on a period of 360 days.

[0037] It should be understood that the various time period division methods can be set by those skilled in the art according to the actual situation, the number of types of time period division and the scope of specific time period division, or according to the experience of those skilled in the art. It should be understood that other time period division methods set by those skilled in the art without creative work all fall within the protection scope of this application.

[0038] Step S102: obtaining a plurality of time series feature labels corresponding to the plurality of remote sensing images according to the plurality of time series image sets and in combination with a preset time series feature model.

[0039] In certain embodiments of the present application, the multiple time series feature labels corresponding to the multiple remote sensing images are obtained based on the multiple groups of time series image sets in combination with a preset time series feature model, specifically including: 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; Iteratively train the time series feature model according to the multiple groups of time series image training sets, input a group of time series image training sets of the corresponding round into the time series feature model for iteration in each round of iteration to adjust 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; The multiple groups of time series image prediction sets are input into the first time series feature model to obtain multiple time series feature labels corresponding to the multiple remote sensing images.

[0040] In certain embodiments of the present application, the preset training classification ratio is preferably set to 3:1, that is, in each group of time series image sets, 75% of the time series image data is a time series image training set, and 25% of the time series image data is a time series image prediction set.

[0041] In certain embodiments of the present application, the timing feature model includes but is not limited to a long short-term memory network (LSTM, Long Short-Term Memory) model, a bidirectional RNN (Bi-RNN, Bi-directional Recurrent Neural Network) model, a deep RNN (Deep RNN, Deep Recurrent Neural Network) model and a temporal convolutional network (TCN, Temporal Convolutional Network) model.

[0042] The present application first divides multiple groups of time series image sets into training sets and prediction sets according to a preset training classification ratio, then performs multiple iterative training on the time series feature model based on the training set, and obtains multiple time series feature labels corresponding to multiple remote sensing images based on the prediction set, which enables the time series feature model to gradually learn the characteristics of changes in the ecological population of coral reefs, thereby being able to extract changes in the ecological population of coral reefs within different time periods and improve recognition precision and accuracy.

[0043] Step S103: Divide the multiple remote sensing images multiple times according to the 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 divided into one type of grid, two types of grid and three types of grid in descending order of grid size; the first image set is divided according to one type of grid, the second image set is divided according to two types of grid, and the third image set is divided according to three types of grid.

[0044] In certain embodiments of the present application, the multiple remote sensing images are divided multiple times according to the preset multiple grid division methods to obtain the first image set, the second image set and the third image set, specifically including: Dividing the plurality of remote sensing images according to the grid size of the first type of grid to obtain a first grid image set; Dividing the plurality of remote sensing images according to the grid sizes of the second type of grids to obtain a second grid image set; Dividing the plurality of remote sensing images according to the grid sizes of the three types of grids 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; The first grid image set, the second grid image set and the third grid image set are subjected to denoising processing, and image feature information is aligned to obtain a first image set, a second image set and a third image set.

[0045] In some embodiments of the present application, the preferred schemes of the first type of grid, the second type of grid and the third type of grid are: Specifically, taking the preferred solution of the above-mentioned type of grid as an example, That is, the current remote sensing image is cut into a grid array of 1280 horizontal columns and 720 vertical rows, 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 after processing the multiple remote sensing images is used as the first grid image set.

[0046] The present application first divides multiple remote sensing images according to the grid sizes of the first and second grids to obtain a first grid image set and a second grid image set, and then divides the multiple remote sensing images according to the grid sizes of the third grid and performs feature enhancement to obtain a third grid image set. This can prevent the third grid image set divided by the grid sizes of the three grids from losing picture feature details, and then denoises the first grid image set, the second grid image set, and the third grid image set and aligns image feature information. This can divide remote sensing images at different granularities, thereby meeting the subsequent need to distinguish coral reef ecological population information at different granularities, and is more adapted to the actual needs of the task.

[0047] Step S104: 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.

[0048] In certain embodiments of the present application, the feature matching is performed 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 including: 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, a plurality of first semantic features, a plurality of second semantic features and a plurality of third semantic features are obtained; 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 tags, multiple second tags, and multiple third tags.

[0049] In certain embodiments of the present application, the feature extraction encoding method specifically includes: Extract features from each image in the current image set to obtain multiple image feature vectors; Semantic recognition coding is performed on the multiple image feature vectors to obtain multiple semantic features.

[0050] In certain embodiments of the present application, the preset ecological feature dictionary includes categories of ecological populations of coral reefs and corresponding features.

[0051] In certain embodiments of the present application, the preferred model on which the semantic recognition coding is based is the LSeg model or the CAT-Seg model.

[0052] In certain embodiments of the present application, the similarity algorithm includes but is not limited to a cosine similarity algorithm, a Pearson correlation coefficient algorithm, and a Jaccard similarity coefficient algorithm.

[0053] The present application first extracts multiple first semantic features, multiple second semantic features and multiple third semantic features through a feature extraction and coding method, and then based on a similarity algorithm, performs feature matching according to a preset ecological feature dictionary to obtain multiple first markers, multiple second markers and multiple third markers. The application can identify the ecological characteristics of coral reefs at different granularities and embed the ecological characteristics into the markers, thereby achieving the purpose of identifying coral reef ecological populations.

[0054] Step S105: According to the multiple first markers and the multiple third markers, the multiple time series feature markers are corrected to obtain multiple corrected feature markers, and according to the multiple corrected feature markers and the multiple second markers, a coral reef ecological identification result of the area to be identified is obtained.

[0055] In certain embodiments of the present application, the correcting the plurality of timing feature marks to obtain a plurality of corrected feature marks according to the plurality of first marks and the plurality of third marks specifically includes: Eliminate the timing feature marks that do not belong to the regions corresponding to the multiple first marks from the multiple timing feature marks to obtain multiple initial corrected feature marks; According to the grid size of the one type of grid corresponding to the multiple first marks, splicing the multiple third marks to obtain multiple grid areas; Initial revised feature marks that do not belong to the multiple grid areas are eliminated from the multiple initial revised feature marks to obtain multiple revised feature marks.

[0056] The present application first eliminates the time series feature markers that do not belong to the corresponding areas of the multiple first markers from the multiple time series feature markers to obtain multiple initial corrected feature markers, can screen the time series feature markers at the corresponding granularity of the first marker, screen out a part of the time series feature markers with identification errors, and then splice the multiple third markers according to the grid size of a type of grid corresponding to the first marker to obtain multiple grid areas, which can simplify the process of screening and eliminating through the third marker and improve the screening efficiency, and then eliminate the initial corrected feature markers that do not belong to the multiple grid areas from the multiple initial corrected feature markers to obtain multiple corrected feature markers, can screen the time series feature markers at the corresponding granularity of the third marker, and further screen out the identification errors in the time series feature markers, thereby improving the recognition accuracy and precision of the ecological populations of coral reefs.

[0057] In certain embodiments of the present application, obtaining the coral reef ecological identification result of the area to be identified based on the multiple corrected characteristic markers and the multiple second markers specifically includes: Obtaining a feature marker intersection set and a feature marker symmetric difference set according to the multiple modified feature markers and the multiple second markers; Selecting feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold value to obtain a feature marker screening set; A plurality of first semantic recognition results corresponding to each feature marker of the feature marker intersection set and a plurality of second semantic recognition results corresponding to each feature marker of the feature marker screening set are obtained, and a coral reef ecological recognition result of the area to be identified is obtained based on the plurality of first semantic recognition results and the plurality of second semantic recognition results.

[0058] The present application first obtains a feature marker intersection and a feature marker symmetric difference set based on multiple corrected feature markers and multiple second markers, and then screens out feature markers in the feature marker symmetric difference set whose identification probability is not greater than a preset identification threshold to obtain a feature marker screening set, which can further eliminate feature markers with identification errors, and then obtains the coral reef ecological identification result of the area to be identified based on the semantic recognition results of the feature marker intersection and the feature marker screening set, which can combine the identification results of different granularities and the identification results with time series characteristics to further improve the identification precision and accuracy of the ecological populations of coral reefs.

[0059] Compared with the prior art, the present application first divides the remote sensing image according to a plurality of time period division methods, and then obtains a plurality of time series feature markers in combination with a time series feature model, which can extract the changes of ecological populations of coral reefs within different time periods, taking into account both the ecological changes within a short time range and the ecological changes within a long time range, thereby improving the recognition precision and accuracy; then the remote sensing image is divided according to a plurality of grid division methods to obtain a first image set, a second image set and a third image set, and then 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 characteristics of coral reefs at different granularities and embed the ecological characteristics into the markers, thereby achieving the purpose of identifying the ecological populations of coral reefs; then, the plurality of time series feature markers are corrected according to the plurality of first markers and the plurality of third markers, and the ecological recognition results of the coral reefs to be identified are obtained according to the plurality of corrected feature markers and the plurality of second markers, which can combine the recognition results of different granularities and the recognition results with time series features, further reduce the recognition error, thereby improving the recognition precision and accuracy of the ecological populations of coral reefs.

[0060] Corresponding to the previous method, see Figure 2 , the embodiment of the present application provides an underwater coral reef ecological diversity identification system, 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 acquisition module 250; The time series division module 210 is used to obtain 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; The time series marking module 220 is used to obtain a plurality of time series feature marks corresponding to the plurality of remote sensing images according to the plurality of time series image sets in combination with a preset time series feature model; The grid division module 230 is used to divide the multiple remote sensing images multiple times according to multiple preset 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 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; The feature matching module 240 is used 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 acquisition module 250 is used to correct the time series feature marks to obtain multiple corrected feature marks according to the multiple first marks and the multiple third marks, and obtain the coral reef ecological identification result of the area to be identified according to the multiple corrected feature marks and the multiple second marks.

[0061] In some embodiments of the present application, the time division module 210 includes an image acquisition submodule 211; the image acquisition submodule includes an image acquisition unit 2111, a feature recognition unit 2112 and a feature fusion unit 2113; The image acquisition unit 2111 is used to acquire a plurality of initial remote sensing images and a plurality of initial infrared images of the area to be identified; The feature recognition unit 2112 is used to obtain common area features of the multiple initial remote sensing images and the multiple initial infrared images based on a preset feature recognition model; The feature fusion unit 2113 is used to fuse the information features of the multiple initial remote sensing images and the multiple initial infrared images according to the common area features to obtain multiple remote sensing images.

[0062] In certain embodiments of the present application, the timing marking module 220 includes a data partitioning submodule 221, a model training submodule 222, and a timing marking submodule 223; The data division submodule 221 is used to divide the multiple groups of time series image sets into multiple groups of time series image training sets and corresponding multiple groups of time series image prediction sets according to a preset training classification ratio; The model training submodule 222 is used to iteratively train the time series feature model according to the multiple groups of time series image training sets, and in each round of iteration, a group of time series image training sets of the corresponding round is input 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 a first time series feature model is obtained and output; The time series marking submodule 223 is used to input the multiple groups of time series image prediction sets into the first time series feature model to obtain multiple time series feature marks corresponding to the multiple remote sensing images.

[0063] In certain embodiments of the present application, the grid division module 230 includes a first division submodule 231, a second division submodule 232, a third division submodule 233 and a feature processing submodule 234; The first division submodule 231 is used to divide the plurality of remote sensing images according to the grid size of the first type of grid to obtain a first grid image set; The second division submodule 232 is used to divide the plurality of remote sensing images according to the grid sizes of the two types of grids to obtain a second grid image set; The third division submodule 233 is used to 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; The feature processing submodule 234 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 a first image set, a second image set and a third image set.

[0064] In certain embodiments of the present application, the feature matching module 240 includes a feature extraction submodule 241 and a feature matching submodule 242; The feature extraction submodule 241 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; The feature matching submodule 242 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.

[0065] In certain embodiments of the present application, the feature extraction encoding method specifically includes: Extract features from each image in the current image set to obtain multiple image feature vectors; Semantic recognition coding is performed on the multiple image feature vectors to obtain multiple semantic features.

[0066] In some embodiments of the present application, the result acquisition module 250 includes a feature correction submodule 251; the feature correction submodule 251 includes an initial correction unit 2511, a grid splicing unit 2512 and a feature correction unit 2513; The initial correction unit 2511 is used to remove the time series feature marks that do not belong to the regions corresponding to the multiple first marks from the multiple time series feature marks, so as to obtain multiple initial correction feature marks; The grid splicing unit 2512 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; The feature correction unit 2513 is used to remove 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.

[0067] In some embodiments of the present application, the result acquisition module 250 includes a result acquisition submodule 252; the result acquisition submodule 252 includes a tag aggregation unit 2521, a tag screening unit 2522 and a result acquisition unit 2523; The tag aggregation unit 2521 is used to obtain a feature tag intersection set and a feature tag symmetric difference set according to the multiple modified feature tags and the multiple second tags; The marker screening unit 2522 is used 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 acquisition unit 2523 is used to obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection set 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 identified based on the multiple first semantic recognition results and the multiple second semantic recognition results.

[0068] The present application first divides the remote sensing image according to a plurality of time period division methods, and then obtains a plurality of time series feature markers in combination with a time series feature model, which can extract the changes of ecological populations of coral reefs within different time periods, taking into account both the ecological changes within a short time range and the ecological changes within a long time range, thereby improving the recognition precision and accuracy; then the remote sensing image is divided according to a plurality of grid division methods to obtain a first image set, a second image set and a third image set, and then 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 characteristics of coral reefs at different granularities and embed the ecological characteristics into the markers, thereby achieving the purpose of identifying the ecological populations of coral reefs; then, the plurality of time series feature markers are corrected according to the plurality of first markers and the plurality of third markers, and the ecological recognition results of the coral reefs to be identified are obtained according to the plurality of corrected feature markers and the plurality of second markers, which can combine the recognition results of different granularities and the recognition results with time series features, further reduce the recognition error, thereby improving the recognition precision and accuracy of the ecological populations of coral reefs.

[0069] Adaptively, the embodiments of the present application also provide a computer device and a computer-readable storage medium.

[0070] The computer device comprises: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor; Wherein, when the processor executes the computer program, an underwater coral reef ecological diversity identification method of the present application is implemented.

[0071] The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute an underwater coral reef ecological diversity identification method of the present application.

[0072] The above are some implementation methods of the present application, and the purpose, technical scheme and beneficial effects of the present application are further described in detail. It should be clear that the above-mentioned partial implementation methods of the present application cannot be understood as limiting the present application. It is particularly pointed out that for those skilled in the art, any changes, modifications, equivalent substitutions and variations made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for identifying underwater coral reef ecological diversity, characterized in that: include: 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; 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; 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; According to the first image set, the second image set and the third image set, 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; According to the multiple first markers and the multiple third markers, the multiple time series feature markers are corrected to obtain multiple corrected feature markers, and according to the multiple corrected feature markers and the multiple second markers, a coral reef ecological identification result of the area to be identified is obtained.

2. A method for identifying underwater coral reef ecological diversity according to claim 1, characterized in that: The step of acquiring a plurality of remote sensing images of the area to be identified specifically includes: Acquire multiple initial remote sensing images and multiple initial infrared images of the area to be identified; Based on a preset feature recognition model, common regional features of the multiple initial remote sensing images and the multiple initial infrared images are obtained; According to the common area features, the multiple initial remote sensing images and the multiple initial infrared images are fused with information features to obtain multiple remote sensing images.

3. The method for identifying underwater coral reef ecological diversity according to claim 1, characterized in that: The step of obtaining a plurality of time series feature tags corresponding to the plurality of remote sensing images based on the plurality of time series image sets and in combination with a preset time series feature model specifically includes: 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; Iteratively train the time series feature model according to the multiple groups of time series image training sets, input a group of time series image training sets of the corresponding round into the time series feature model for iteration in each round of iteration to adjust 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; The multiple groups of time series image prediction sets are input into the first time series feature model to obtain multiple time series feature labels corresponding to the multiple remote sensing images.

4. The method for identifying underwater coral reef ecological diversity according to claim 1, characterized in that: The step of dividing the plurality of remote sensing images multiple times according to the preset plurality of grid division methods to obtain the first image set, the second image set and the third image set specifically includes: Dividing the plurality of remote sensing images according to the grid size of the first type of grid to obtain a first grid image set; Dividing the plurality of remote sensing images according to the grid sizes of the second type of grids to obtain a second grid image set; Dividing the plurality of remote sensing images according to the grid sizes of the three types of grids 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; The first grid image set, the second grid image set and the third grid image set are subjected to denoising processing, and image feature information is aligned to obtain a first image set, a second image set and a third image set.

5. The method for identifying underwater coral reef ecological diversity according to claim 1, characterized in that: The method of 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 a plurality of first markers, a plurality of second markers and a plurality of third markers specifically includes: 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, a plurality of first semantic features, a plurality of second semantic features and a plurality of third semantic features are obtained; 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 tags, multiple second tags, and multiple third tags.

6. A method for identifying underwater coral reef ecological diversity according to claim 5, characterized in that: The feature extraction and encoding method specifically comprises: Extract features from each image in the current image set to obtain multiple image feature vectors; Semantic recognition coding is performed on the multiple image feature vectors to obtain multiple semantic features.

7. The method for identifying underwater coral reef ecological diversity according to claim 1, characterized in that: The step of correcting the plurality of timing feature marks to obtain a plurality of corrected feature marks according to the plurality of first marks and the plurality of third marks specifically includes: Eliminate the timing feature marks that do not belong to the regions corresponding to the multiple first marks from the multiple timing feature marks to obtain multiple initial corrected feature marks; According to the grid size of the one type of grid corresponding to the multiple first marks, splicing the multiple third marks to obtain multiple grid areas; Initial revised feature marks that do not belong to the multiple grid areas are eliminated from the multiple initial revised feature marks to obtain multiple revised feature marks.

8. The method for identifying underwater coral reef ecological diversity according to claim 1, characterized in that: The obtaining of the coral reef ecological identification result of the area to be identified based on the multiple corrected characteristic markers and the multiple second markers specifically includes: Obtaining a feature marker intersection set and a feature marker symmetric difference set according to the multiple modified feature markers and the multiple second markers; Selecting feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold value to obtain a feature marker screening set; A plurality of first semantic recognition results corresponding to each feature marker of the feature marker intersection set and a plurality of second semantic recognition results corresponding to each feature marker of the feature marker screening set are obtained, and a coral reef ecological recognition result of the area to be identified is obtained based on the plurality of first semantic recognition results and the plurality of second semantic recognition results.

9. An underwater coral reef ecological diversity identification 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 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; The time series marking module is used to obtain a plurality of time series feature marks corresponding to the plurality of remote sensing images according to the plurality of 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 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; The feature matching module is used 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 acquisition module is used to correct the time series feature marks to obtain multiple corrected feature marks according to the multiple first marks and the multiple third marks, and obtain the coral reef ecological identification result of the area to be identified according to the multiple corrected feature marks and the multiple second marks.

10. The underwater coral reef ecological diversity identification system according to claim 9, characterized in that: 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; The tag aggregation unit is used to obtain a feature tag intersection set and a feature tag symmetric difference set according to the multiple modified feature tags and the multiple second tags; The marker screening unit is used to screen the feature markers in the feature marker symmetric difference set whose corresponding recognition probabilities are not greater than a preset recognition threshold value, to obtain a feature marker screening set; The result acquisition unit is used to obtain multiple first semantic recognition results corresponding to each feature marker in the feature marker intersection set 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 identified based on the multiple first semantic recognition results and the multiple second semantic recognition results.

Images