Marine organism type prediction method and system based on deep learning

Abstract

The invention discloses a marine organism type prediction method and system based on deep learning, and relates to the field of marine organism detection.The marine organism type prediction method comprises the steps that a basic graph and a line library are constructed, and the basic graph and the line library comprise basic elements; performing image processing on the input marine organism image, and decomposing the input marine organism image into a combination of a basic graph and lines; distributing a unique identifier for each basic graph and line in the basic graph and line library, and presetting a combination rule of the basic graphs and lines; collecting a large amount of marked marine organism image data, respectively carrying out image processing, extracting combined features of basic graphs and lines, and carrying out deep learning model training by using the combined features; performing image decomposition and feature extraction on a marine organism image to be identified, and inputting image decomposition and feature extraction results into the trained deep learning model for classification and identification; the recognition precision of the marine organism image can be improved.

Description

Technical Field

[0001] The present invention relates to the field of marine life detection, and in particular to a method and system for predicting marine life types based on deep learning. Background Art

[0002] Marine life is an important part of the earth's ecosystem, and its diversity and distribution have a significant impact on the global ecological environment. Due to the complexity and unpredictability of the marine environment, the study and monitoring of marine life types has always been a challenge in marine scientific research; with the continuous deepening of marine research, marine biodiversity monitoring and research has become an important part of marine science; the classification and identification of marine life is a basic task in marine ecosystem research, and is of great significance for the protection and rational use of marine resources; traditional marine life identification methods mainly rely on manual observation and laboratory analysis, which not only consumes a lot of manpower and material resources, but also has problems such as limited sampling range and long time period; these methods are not only time-consuming and labor-intensive, but also require high professional knowledge, and it is difficult to meet the needs of large-scale marine life monitoring.

[0003] In the prior art, a Chinese invention patent with announcement number CN117911793B discloses a method for intelligent detection of marine life based on deep learning, including: collecting different marine life image data from a public database, the marine life image data including image data of different species, environments and angles, and receiving the marine life image data to be detected sent by the user, using a convolutional neural network algorithm to construct a model according to different marine life image data, and the constructed model detects the marine life species and predicted probability of the marine life image data to be detected, using a generative adversarial network algorithm to construct a model according to different marine life image data, and the constructed model detects the marine life species and predicted probability of the marine life image data to be detected, comparing the detection results of the convolutional neural network algorithm and the generative adversarial network algorithm, and determining the final result and optimizing the processing according to the consistency of the detection results; However, the above invention still has the following problems: the marine life in the marine life image data to be detected has multiple forms and can change dynamically. When the convolutional neural network algorithm and the generative adversarial network algorithm are used for identification, there is a problem of inaccurate identification. Summary of the invention

[0004] The embodiments of the present application provide a method and system for predicting marine life types based on deep learning, thereby solving the problem in the prior art that marine life in the marine life image data to be detected has various forms and can change dynamically, and inaccurate recognition occurs when using convolutional neural network algorithms and generative adversarial network algorithms for recognition, thereby improving the accuracy of model recognition.

[0005] The present application embodiment provides a method for predicting marine organism types based on deep learning, including:

[0006] S1: constructing a basic graphics and line library, wherein the basic graphics and line library includes basic elements, and the basic elements can cover the features of the appearance of marine organisms;

[0007] The basic shapes include but are not limited to circles, ellipses, rectangles, triangles, polygons, hemispheres, frustums of different tapers, and cones;

[0008] The lines include but are not limited to straight lines, curves (such as sine waves, cosine waves), spiral lines, etc., as well as lines of different thicknesses and color variations;

[0009] S2: performing image processing on the input marine life image to decompose the input marine life image into a combination of basic graphics and lines, specifically comprising the following steps:

[0010] By using an edge detection algorithm to identify the boundary line in the marine life image, matching the corresponding graphics in the basic graphics library according to the shape and characteristics of the boundary line, a first basic graphic is obtained;

[0011] For the complex parts that cannot be directly matched to the basic figures, lines are used to express them and the first lines are obtained;

[0012] S3: assigning a unique identifier to each basic graphic and line in the basic graphic and line library, and presetting a combination rule of the basic graphics and lines;

[0013] The identifier is a code, which consists of numbers and letters. The code of each basic figure and line is unique and can be checked;

[0014] The combination rules represent the logical relationship of the appearance structure of marine organisms, such as relative position, angle, size ratio, etc.

[0015] For example, the tentacles of an octopus are made up of multiple cone-shaped basic shapes connected by spiral lines, and the connection between the hemispherical top of the head and the body is smoothly transitioned by a curve;

[0016] S4: Collect a large amount of labeled marine biological image data, perform image processing respectively, extract the combined features of the first basic image and the first line, and use the combined features to train the deep learning model;

[0017] The deep learning model is a convolutional neural network model or a generative adversarial network model;

[0018] During the model training process, learn the recognition of individual basic elements, the combination of basic elements and the logical relationship;

[0019] S5: Decompose the image of the marine organism to be identified and perform feature extraction, and input the image decomposition and feature extraction results into the trained deep learning model for classification and identification;

[0020] The deep learning model determines which marine organism the image belongs to based on the feature combination rules learned during the training process.

[0021] In some embodiments, in step S3, the combination rule of the basic graphics and lines is the combination rule of the coding, which is recorded as the coding logic module, and there is a module with dynamic change capability in the coding logic module, which is recorded as the dynamic logic module;

[0022] The dynamic logic module includes a combination of basic graphics and lines with dynamic change characteristics, and the variable factors in the dynamic change characteristics are specifically:

[0023] Combination ratio: the ratio of different basic shapes and lines when combined;

[0024] For example, the sizes and proportions of the various parts of images taken of different octopuses or the same octopus at different growth stages are different. The dynamic logic module recognizes and adjusts the proportion changes when basic graphics and lines are combined;

[0025] Connection mode: the connection method between basic elements, such as straight line connection, curve connection, rotation connection, etc.;

[0026] Spatial position: the position of the basic element in the assembly;

[0027] Due to different shooting angles, the relative positions and angles of the basic graphics and lines in the marine life images will also change. The dynamic logic module can adjust the relative positions and angles.

[0028] Generate images of marine life when variable factors change through algorithm simulation, input the images of marine life into the deep learning model, and perform model training so that deep learning can learn the changes in proportion, connection mode and spatial position of marine life when variable factors change;

[0029] When the image of marine organisms to be detected is input into the deep learning model, the deep learning model automatically adjusts the position and connection method of the basic elements according to the learned proportion changes, connection methods and spatial positions of the marine organisms, and outputs the recognition results.

[0030] In some embodiments, the dynamic logic module is differentiated to obtain a ratio adjustment submodule, a connection mode optimization submodule and a spatial position adjustment submodule, and the specific functions of each module are as follows:

[0031] Proportional adjustment submodule: dynamically adjust the position and size of other basic elements according to the proportion changes of basic shapes and lines in the combination;

[0032] Connection mode optimization submodule: optimizes the connection mode and changes of the overall structure according to the changes in the connection mode between basic elements;

[0033] Spatial position adjustment submodule: dynamically adjusts the image layout according to the position changes of the basic elements in space to adapt to marine life of different forms and postures;

[0034] By analyzing the dynamic changes of the proportion adjustment submodule, the connection mode optimization submodule and the spatial position adjustment submodule, the variable factors of each submodule are obtained. The variable factors of each submodule are:

[0035] Variable factors of the proportion adjustment submodule: including the proportion coefficients of each basic figure and line, the balance index of the overall structure, etc.

[0036] Variable factors of the connection mode optimization submodule: including the type of connection line (straight line, curve, spiral line, etc.), the location of the connection point, the connection angle, etc.;

[0037] Variable factors of the spatial position adjustment submodule: including the coordinates of each basic element in two-dimensional or three-dimensional space, the relative distance and direction between elements, etc.;

[0038] Generate images of marine life when the variable factors of each submodule change through algorithm simulation, and simulate the dynamic adjustment of each submodule when the variable factors of each submodule change;

[0039] For example, when the length of the tentacle increases, the proportion adjustment submodule automatically adjusts the proportion of the tentacle to the rest of the body. The proportion adjustment submodule recalculates the position of the end of the tentacle to ensure that the entire image remains proportional.

[0040] In some embodiments, the variable factors of the spatial position adjustment submodule are the coordinates of each basic element in a two-dimensional or three-dimensional space, the relative distance and direction between the elements. The variable factors of the spatial position adjustment submodule are distinguished, which can be specifically distinguished as follows:

[0041] Coordinate adjustment variables: the position of the basic element in the coordinate system, such as x, y coordinates or x, y, z coordinates;

[0042] Distance adjustment variable: the distance between basic elements to ensure that they do not block or overlap each other;

[0043] Orientation Correction Variable: Adjusts the orientation of the element to align with other base elements in the marine life image;

[0044] By analyzing the dynamic changes of coordinate adjustment variables, distance adjustment variables and direction correction variables, the analysis results are as follows:

[0045] Dynamic changes of coordinate adjustment variables: when the positions of other basic elements in the image change, the coordinate adjustment variables automatically adjust the new coordinates of the basic elements to keep the relative positions between elements and the overall structure reasonable;

[0046] Dynamic changes of the distance adjustment variable: when it is detected that the distance between basic elements is too close, resulting in occlusion, the distance adjustment variable increases the distance between basic elements; conversely, when it is detected that the distance between elements is too far, resulting in weakened correlation, the distance adjustment variable decreases the distance between basic elements;

[0047] Dynamic changes of direction correction variables: The direction correction variables dynamically correct the direction of elements according to the overall layout of the image and the logical position of the basic elements to ensure the logical coherence and recognition accuracy of the image.

[0048] The spatial position adjustment submodule dynamically adjusts the spatial position of the element when the coordinate adjustment variable, distance adjustment variable and direction correction variable are dynamically changed through algorithm simulation;

[0049] For example, when a fish changes its swimming direction, the spatial position regulation submodule detects the change and automatically adjusts the direction of the fish's head and body to align with the new swimming direction.

[0050] In order to better demonstrate a method for predicting marine life types based on deep learning, the present invention provides a marine life type prediction system based on deep learning, the system comprising the following steps:

[0051] Building a basic graphics and line library module: Building a basic graphics and line library, the basic graphics and line library contains basic elements, and the basic elements can cover the characteristics of the appearance of marine organisms;

[0052] Image processing module: performs image processing on the input marine life image, and decomposes the input marine life image into a combination of basic graphics and lines;

[0053] Coding combination rule module: assigns a unique identifier to each basic figure and line in the basic figure and line library, and presets the combination rules of the basic figures and lines;

[0054] Model training module: collect a large amount of annotated marine biological image data, perform image processing respectively, extract the combined features of the first basic image and the first line, and use the combined features to train the deep learning model;

[0055] Model recognition module: Perform image decomposition and feature extraction on the marine organism images to be identified, and input the image decomposition and feature extraction results into the trained deep learning model for classification and recognition.

[0056] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0057] By constructing a library of basic graphics and lines, the features of the appearance of marine life can be covered, making the image decomposition and feature extraction process more efficient and accurate; by introducing a deep learning model, the classification and recognition capabilities of complex marine life images are improved, and the type of marine life in the image can be more accurately judged; through deep learning model training with a large amount of labeled marine life image data, the model can learn the general laws and feature combination methods of the appearance structure of marine life, so that the model can maintain a high recognition accuracy when facing new, unlabeled marine life images, that is, it has a strong generalization ability.

[0058] The design of the basic graphics and line library makes this solution flexible. As our understanding of marine life deepens, we can continue to add new basic elements to the library or adjust the parameters of existing elements to accommodate a wider range of marine life types. The above scalability ensures the long-term effectiveness and adaptability of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of a method and system for predicting marine organism types based on deep learning according to the present invention. DETAILED DESCRIPTION

[0060] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0061] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are only for illustrative purposes and do not represent the only implementation method.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0063] Embodiment 1: Figure 1 As shown, the present application provides a method for predicting marine organism types based on deep learning, comprising:

[0064] S1: constructing a basic graphics and line library, wherein the basic graphics and line library includes basic elements, and the basic elements can cover the features of the appearance of marine organisms;

[0065] The basic shapes include but are not limited to circles, ellipses, rectangles, triangles, polygons, hemispheres, frustums of different tapers, and cones;

[0066] The lines include but are not limited to straight lines, curves (such as sine waves, cosine waves), spiral lines, etc., as well as lines of different thicknesses and color variations;

[0067] S2: performing image processing on the input marine life image to decompose the input marine life image into a combination of basic graphics and lines, specifically comprising the following steps:

[0068] By using an edge detection algorithm to identify the boundary line in the marine life image, matching the corresponding graphics in the basic graphics library according to the shape and characteristics of the boundary line, a first basic graphic is obtained;

[0069] For the complex parts that cannot be directly matched to the basic figures, lines are used to express them and the first lines are obtained;

[0070] S3: assigning a unique identifier to each basic graphic and line in the basic graphic and line library, and presetting a combination rule of the basic graphics and lines;

[0071] The identifier is a code, which consists of numbers and letters. The code of each basic figure and line is unique and can be checked;

[0072] The combination rules represent the logical relationship of the appearance structure of marine organisms, such as relative position, angle, size ratio, etc.

[0073] For example, the tentacles of an octopus are made up of multiple cone-shaped basic shapes connected by spiral lines, and the connection between the hemispherical top of the head and the body is smoothly transitioned by a curve;

[0074] S4: Collect a large amount of labeled marine biological image data, perform image processing respectively, extract the combined features of the first basic image and the first line, and use the combined features to train the deep learning model;

[0075] The deep learning model is a convolutional neural network model or a generative adversarial network model;

[0076] During the model training process, learn the recognition of individual basic elements, the combination of basic elements and the logical relationship;

[0077] S5: Decompose the image of the marine organism to be identified and perform feature extraction, and input the image decomposition and feature extraction results into the trained deep learning model for classification and identification;

[0078] The deep learning model determines which marine organism the image belongs to based on the feature combination rules learned during the training process.

[0079] In some embodiments, the basic graphics and line library is optimized according to the recognition results, and new basic elements are added or parameters of existing elements are adjusted.

[0080] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0081] By constructing a library of basic graphics and lines, the features of the appearance of marine life can be covered, making the image decomposition and feature extraction process more efficient and accurate; by introducing a deep learning model, the classification and recognition capabilities of complex marine life images are improved, and the type of marine life in the image can be more accurately judged; through deep learning model training with a large amount of labeled marine life image data, the model can learn the general laws and feature combination methods of the appearance structure of marine life, so that the model can maintain a high recognition accuracy when facing new, unlabeled marine life images, that is, it has a strong generalization ability.

[0082] The design of the basic graphics and line library makes this solution flexible. As our understanding of marine life deepens, we can continue to add new basic elements to the library or adjust the parameters of existing elements to accommodate a wider range of marine life types. The above scalability ensures the long-term effectiveness and adaptability of the solution.

[0083] Embodiment 2: In embodiment 1, the morphologies of different marine organisms of the same species in the marine organism images to be identified are greatly different, and even the images of the same marine organism taken from different shooting angles are different, resulting in a problem of reduced recognition accuracy when the marine organism images to be identified are detected.

[0084] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0085] In some embodiments, in step S3, the combination rule of the basic graphics and lines is the combination rule of the coding, which is recorded as the coding logic module, and there is a module with dynamic change capability in the coding logic module, which is recorded as the dynamic logic module;

[0086] The dynamic logic module includes a combination of basic graphics and lines with dynamic change characteristics, and the variable factors in the dynamic change characteristics are specifically:

[0087] Combination ratio: the ratio of different basic shapes and lines when combined;

[0088] For example, the sizes and proportions of the various parts of images taken of different octopuses or the same octopus at different growth stages are different. The dynamic logic module recognizes and adjusts the proportion changes when basic graphics and lines are combined;

[0089] Connection mode: the connection method between basic elements, such as straight line connection, curve connection, rotation connection, etc.;

[0090] Spatial position: the position of the basic element in the assembly;

[0091] Due to different shooting angles, the relative positions and angles of the basic graphics and lines in the marine life images will also change. The dynamic logic module can adjust the relative positions and angles.

[0092] Generate images of marine life when variable factors change through algorithm simulation, input the images of marine life into the deep learning model, and perform model training so that deep learning can learn the changes in proportion, connection mode and spatial position of marine life when variable factors change;

[0093] When the image of marine organisms to be detected is input into the deep learning model, the deep learning model automatically adjusts the position and connection method of the basic elements according to the learned proportion changes, connection methods and spatial positions of the marine organisms, and outputs the recognition results.

[0094] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0095] By introducing a dynamic logic module, this module can handle the recognition difficulties caused by morphological differences and changes in shooting angles in marine life images, and dynamically adjust the combination ratio, connection pattern and spatial position of basic graphics and lines, so that the model can accurately identify marine life in different forms and angles, thereby improving recognition accuracy.

[0096] Traditional recognition methods often rely on fixed templates or rules, and are difficult to cope with images of marine life with varied morphologies and shooting angles. The technical solution in this application generates images of marine life when various variable factors change through algorithm simulation, and performs model training, so that the deep learning model can learn these changing rules, thereby enhancing the generalization ability of the model and maintaining a high recognition accuracy in different scenarios.

[0097] Embodiment 3: In Embodiment 1 and Embodiment 2, on the basis of the dynamic logic module, the dynamic logic module is subdivided to distinguish different sub-modules, so that the dynamically changing logic can be accurately controlled.

[0098] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0099] In some embodiments, the dynamic logic module is differentiated to obtain a ratio adjustment submodule, a connection mode optimization submodule and a spatial position adjustment submodule, and the specific functions of each module are as follows:

[0100] Proportional adjustment submodule: dynamically adjust the position and size of other basic elements according to the proportion changes of basic shapes and lines in the combination;

[0101] Connection mode optimization submodule: optimizes the connection mode and changes of the overall structure according to the changes in the connection mode between basic elements;

[0102] Spatial position adjustment submodule: dynamically adjusts the image layout according to the position changes of the basic elements in space to adapt to marine life of different forms and postures;

[0103] By analyzing the dynamic changes of the proportion adjustment submodule, the connection mode optimization submodule and the spatial position adjustment submodule, the variable factors of each submodule are obtained. The variable factors of each submodule are:

[0104] Variable factors of the proportion adjustment submodule: including the proportion coefficients of each basic figure and line, the balance index of the overall structure, etc.

[0105] Variable factors of the connection mode optimization submodule: including the type of connection line (straight line, curve, spiral line, etc.), the location of the connection point, the connection angle, etc.;

[0106] Variable factors of the spatial position adjustment submodule: including the coordinates of each basic element in two-dimensional or three-dimensional space, the relative distance and direction between elements, etc.;

[0107] Generate images of marine life when the variable factors of each submodule change through algorithm simulation, and simulate the dynamic adjustment of each submodule when the variable factors of each submodule change;

[0108] For example, when the length of the tentacle increases, the proportion adjustment submodule automatically adjusts the proportion of the tentacle to the rest of the body. The proportion adjustment submodule recalculates the position of the end of the tentacle to ensure that the entire image remains proportional.

[0109] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0110] By subdividing the dynamic logic module, we obtain the proportion adjustment submodule, the connection mode optimization submodule and the spatial position adjustment submodule, thereby achieving refined control over each detail in the graphic combination; subdivision enables the model to accurately respond to and adapt to changes in basic elements, thereby improving the accuracy and flexibility of the generated image.

[0111] Each sub-module dynamically adjusts the position, size and connection method of other elements according to the changes of the basic elements, so that the entire image can automatically adapt to different forms and postures. The model can accurately identify marine life in different forms and angles, thereby improving recognition accuracy.

[0112] Embodiment 4: In the above embodiment, the spatial position adjustment submodule includes multiple variable factors. The variable factors of the spatial position adjustment submodule are subdivided to distinguish different sub-variable factors, so that the logic of dynamic changes can be accurately controlled.

[0113] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0114] In some embodiments, the variable factors of the spatial position adjustment submodule are the coordinates of each basic element in a two-dimensional or three-dimensional space, the relative distance and direction between the elements. The variable factors of the spatial position adjustment submodule are distinguished, which can be specifically distinguished as follows:

[0115] Coordinate adjustment variables: the position of the basic element in the coordinate system, such as x, y coordinates or x, y, z coordinates;

[0116] Distance adjustment variable: the distance between basic elements to ensure that they do not block or overlap each other;

[0117] Orientation Correction Variable: Adjusts the orientation of the element to align with other base elements in the marine life image;

[0118] By analyzing the dynamic changes of coordinate adjustment variables, distance adjustment variables and direction correction variables, the analysis results are as follows:

[0119] Dynamic changes of coordinate adjustment variables: when the positions of other basic elements in the image change, the coordinate adjustment variables automatically adjust the new coordinates of the basic elements to keep the relative positions between elements and the overall structure reasonable;

[0120] Dynamic changes of the distance adjustment variable: when it is detected that the distance between basic elements is too close, resulting in occlusion, the distance adjustment variable increases the distance between basic elements; conversely, when it is detected that the distance between elements is too far, resulting in weakened correlation, the distance adjustment variable decreases the distance between basic elements;

[0121] Dynamic changes of direction correction variables: The direction correction variables dynamically correct the direction of elements according to the overall layout of the image and the logical position of the basic elements to ensure the logical coherence and recognition accuracy of the image.

[0122] The spatial position adjustment submodule dynamically adjusts the spatial position of the element when the coordinate adjustment variable, distance adjustment variable and direction correction variable are dynamically changed through algorithm simulation;

[0123] For example, when a fish changes its swimming direction, the spatial position regulation submodule detects the change and automatically adjusts the direction of the fish's head and body to align with the new swimming direction.

[0124] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0125] By subdividing the variable factors of the spatial position adjustment submodule and introducing coordinate adjustment variables, distance adjustment variables and direction correction variables, the position, distance and direction of each basic element in two-dimensional or three-dimensional space can be accurately controlled; the flexibility and accuracy of the system are improved through subdivision and dynamic adjustment methods, and it can adapt to complex image generation.

[0126] By analyzing the dynamic changes of variable factors and adjusting the spatial position of elements accordingly, the system can respond in real time to changes in elements in the image, such as position movement and direction change. This real-time performance ensures that the generated image has a high degree of coherence and naturalness, thereby improving recognition accuracy.

[0127] In order to better demonstrate a method for predicting marine life types based on deep learning, the present invention provides a system for predicting marine life types based on deep learning, the system comprising the following steps:

[0128] Building a basic graphics and line library module: Building a basic graphics and line library, the basic graphics and line library contains basic elements, and the basic elements can cover the characteristics of the appearance of marine organisms;

[0129] Image processing module: performs image processing on the input marine life image, and decomposes the input marine life image into a combination of basic graphics and lines;

[0130] Coding combination rule module: assigns a unique identifier to each basic figure and line in the basic figure and line library, and presets the combination rules of the basic figures and lines;

[0131] Model training module: collect a large amount of annotated marine biological image data, perform image processing respectively, extract the combined features of the first basic image and the first line, and use the combined features to train the deep learning model;

[0132] Model recognition module: Perform image decomposition and feature extraction on the marine organism images to be identified, and input the image decomposition and feature extraction results into the trained deep learning model for classification and recognition.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting marine organism types based on deep learning, characterized in that: include: S1: constructing a basic graphics and line library, wherein the basic graphics and line library includes basic elements, and the basic elements can cover the features of the appearance of marine organisms; S2: Collect a large number of marine life images, perform image processing on the marine life images, and decompose the marine life images into a combination of basic graphics and lines; S3: assigning a unique identifier to each basic figure and line in the basic figure and line library, and presetting the combination rules of the basic figures and lines; S4: Collect a large amount of annotated marine biological image data, perform image processing, extract the combined features of basic graphics and lines, and use the combined features of basic graphics and lines to train deep learning models; S5: Decompose the image of the marine organism to be identified and perform feature extraction, and input the image decomposition and feature extraction results into the trained deep learning model for classification and identification.

2. A method for predicting marine life types based on deep learning as claimed in claim 1, characterized in that: The image processing of the marine life image is performed to decompose the input marine life image into a combination of basic graphics and lines, which specifically includes the following steps: By using an edge detection algorithm to identify the boundary line in the marine life image, matching the corresponding graphics in the basic graphics library according to the shape and characteristics of the boundary line, a first basic graphic is obtained; For complex parts that cannot be directly matched to the basic shape, lines are used to connect them to obtain the first lines.

3. A method for predicting marine life types based on deep learning as claimed in claim 1, characterized in that: The deep learning model is a convolutional neural network model or a generative adversarial network model.

4. A method for predicting marine life types based on deep learning as claimed in claim 1, characterized in that: The combination rules of the basic graphics and lines are recorded as encoding logic modules.

5. A method for predicting marine life types based on deep learning as claimed in claim 4, characterized in that: The encoding logic module includes a module with dynamic change capability, which is referred to as a dynamic logic module.

6. A method for predicting marine life types based on deep learning as claimed in claim 5, characterized in that: The dynamic logic module includes a combination of basic graphics and lines with dynamic change characteristics, and the variable factors in the dynamic change characteristics are specifically: Combination ratio: the ratio of different basic shapes and lines when combined; Connection mode: the connection method between basic elements; Spatial position: the position of the basic element in the assembly; Generate images of marine life when variable factors change through algorithm simulation, input the images of marine life into the deep learning model, and perform model training so that deep learning can learn the changes in proportion, connection mode and spatial position of marine life when variable factors change; When the image of marine organisms to be detected is input into the deep learning model, the deep learning model automatically adjusts the position and connection method of the basic elements according to the learned proportion changes, connection methods and spatial positions of the marine organisms, and outputs the recognition results.

7. A method for predicting marine life types based on deep learning as claimed in claim 6, characterized in that: The dynamic logic module is differentiated to obtain a proportion adjustment submodule, a connection mode optimization submodule and a spatial position adjustment submodule. The specific functions of each module are as follows: Proportional adjustment submodule: dynamically adjust the position and size of other basic elements according to the proportion changes of basic shapes and lines in the combination; Connection mode optimization submodule: optimizes the connection mode and changes of the overall structure according to the changes in the connection mode between basic elements; Spatial position adjustment submodule: dynamically adjust the image layout according to the position changes of basic elements in space.

8. A method for predicting marine life types based on deep learning as claimed in claim 7, characterized in that: By analyzing the dynamic changes of the proportion adjustment submodule, the connection mode optimization submodule and the spatial position adjustment submodule, the variable factors of each submodule are obtained. The variable factors of each submodule are: Variable factors of the proportion adjustment submodule: including the proportion coefficients of each basic figure and line and the balance index of the overall structure; Variable factors of the connection mode optimization submodule: including the type of connection line, the location of the connection point, and the connection angle; Variable factors of the spatial position adjustment submodule: including the coordinates of each basic element in two-dimensional or three-dimensional space, the distance and direction between basic elements; The algorithm is used to simulate and generate images of marine life when the variable factors of each submodule change, and to simulate the dynamic adjustment of each submodule when the variable factors of each submodule change.

9. A method for predicting marine life types based on deep learning as claimed in claim 8, characterized in that: The variable factors of the spatial position adjustment submodule are the coordinates of each basic element in a two-dimensional or three-dimensional space, the distance and direction between the basic elements. The variable factors of the spatial position adjustment submodule are specifically divided into: Coordinate adjustment variables: the position of the basic element in the coordinate system; Distance moderator variable: distance between basic elements; Orientation Correction Variable: Adjusts the orientation of the base element to be consistent with other base elements in the marine life image; By analyzing the dynamic changes of coordinate adjustment variables, distance adjustment variables and direction correction variables, the analysis results are as follows: Dynamic changes of coordinate adjustment variables: when the positions of other basic elements in the image change, the coordinate adjustment variables automatically adjust the new coordinates of the basic elements; Dynamic changes of the distance adjustment variable: when it is detected that the distance between basic elements is too close, resulting in occlusion, the distance adjustment variable increases the distance between basic elements; conversely, when it is detected that the distance between elements is too far, resulting in weakened correlation, the distance adjustment variable decreases the distance between basic elements; Dynamic changes of direction correction variables: Direction correction variables dynamically correct the direction of elements according to the overall layout of the image and the logical position of the basic elements; The spatial position adjustment submodule dynamically adjusts the spatial position of the element when the coordinate adjustment variable, distance adjustment variable and direction correction variable change dynamically through algorithm simulation.

10. A deep learning-based marine life type prediction system, using the deep learning-based marine life type prediction method according to claim 1, characterized in that: include: Building a basic graphics and line library module: Building a basic graphics and line library, the basic graphics and line library contains basic elements, and the basic elements can cover the characteristics of the appearance of marine organisms; Image processing module: performs image processing on the input marine life image, and decomposes the input marine life image into a combination of basic graphics and lines; Coding combination rule module: assigns a unique identifier to each basic figure and line in the basic figure and line library, and presets the combination rules of the basic figures and lines; Model training module: collect a large amount of annotated marine biological image data, perform image processing respectively, extract the combined features of the first basic image and the first line, and use the combined features to train the deep learning model; Model recognition module: Perform image decomposition and feature extraction on the marine organism images to be identified, and input the image decomposition and feature extraction results into the trained deep learning model for classification and recognition.

Images