Living fish underwater feature database construction method and system, and medium

Abstract

The invention discloses a method and system for constructing a living fish underwater feature database and a medium, and the method comprises the steps: collecting original fish information through a preset frequency, carrying out the three-dimensional point cloud reconstruction of a second fish image obtained after preprocessing, and extracting the static phenotypic parameters, such as the scale arrangement features and fin ray morphological features, of the surface of a fish body; body surface texture reflection characteristics are obtained through multispectral analysis; meanwhile, associating time domain features of the second fish image and the second acoustic signal, extracting dynamic behavior features such as a motion trail and swimming bladder vibration frequency, and establishing a feature coding library containing typical behavior patterns; a mapping relation network of different disease types, body surface lesion area spatial distribution and behavior abnormal modes is constructed based on the features, and a feature parameter conversion relation is established; finally, various characteristic parameters are stored in a structured mode, a multi-dimensional characteristic database is formed, multi-dimensional correlation analysis of static phenotypes and dynamic behaviors of fishes is achieved, and the accuracy of disease diagnosis and cross-species characteristic migration is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a method and system for constructing an underwater feature database of live fish and a medium. BACKGROUND

[0002] Existing fish feature databases are mostly constructed based on single modal data, which can associate diseases with phenotypic or behavioral characteristics, but have the following limitations: multi-source data are not time-space aligned and co-analyzed, resulting in disease diagnosis based on a single factor; the behavioral feature library only records isolated time series data, lacking coded expression of typical behavior patterns; and the association network between body surface lesions and behavioral abnormalities has not been systematically established. SUMMARY

[0003] In view of the above problems, the present application provides a method and system for constructing an underwater feature database of live fish and a medium, which solves the problem of insufficient disease identification capability of fish feature databases.

[0004] To achieve the above-mentioned purpose, in a first aspect, the present application provides a method for constructing an underwater feature database of live fish, comprising:

[0005] Collecting original fish information at a preset frequency, the original fish information including first fish images, first acoustic information and first environmental parameters;

[0006] Pretreating the original fish information to obtain processed fish information, the pretreatment including data cleaning and time-space alignment, the processed fish information including second fish images, second acoustic information and second environmental parameters after time-space alignment;

[0007] Reconstructing the second fish images in three-dimensional point cloud to extract static phenotypic parameters, the static phenotypic parameters including fish body surface scale arrangement features and fin strip morphological features, and performing multi-spectral analysis on the second fish images to obtain body surface texture reflection characteristics;

[0008] Correlating the second fish images and the second acoustic signals in time domain to extract dynamic behavior features, the dynamic behavior features including fish body movement trajectory and swim bladder vibration frequency, and establishing a behavior feature code library containing typical behavior patterns, the typical behavior patterns including at least one of feeding, escaping and fighting;

[0009] According to the static phenotypic parameters, the body surface texture reflection characteristics, the dynamic behavior features and the behavior feature code library, a mapping relationship network of different disease types and body surface lesion region spatial distribution, behavior abnormality patterns is constructed;

[0010] Based on the static phenotypic parameters and the dynamic behavior features, a feature parameter conversion relationship between a reference fish species and a target fish species is established;

[0011] The static phenotype parameters, dynamic behavior characteristics, behavior characteristic code library, mapping relationship network and characteristic parameter conversion relationship are stored in a structured manner to build a multi-dimensional feature database containing fish phenotype characteristics, behavior characteristics, disease characteristics and environmental parameters.

[0012] The original fish information collected at the preset frequency can cover various economic fish species, such as large yellow croaker, orange bass, grouper and eel, and the application scenarios include but are not limited to:

[0013] Intensive culture pond: fish behavior in high-density culture environment is monitored by fixed underwater camera equipment and acoustic sensors;

[0014] Net cage culture: mobile collection devices are used to obtain the dynamic characteristics of fish in the net cage;

[0015] Water-based roadbed culture: multispectral imaging technology is used to analyze the phenotype and pathological characteristics of fish in the roadbed pond.

[0016] In some embodiments, the original fish information is preprocessed to obtain processed fish information, and the preprocessing includes data cleaning and spatio-temporal alignment. The processed fish information includes the second fish image, the second acoustic information and the second environmental parameter after spatio-temporal alignment, and the second fish image, the second acoustic information and the second environmental parameter include:

[0017] The images meeting the preset clarity threshold are selected from the first fish image as the second fish image, and the preset clarity threshold includes edge sharpness and contrast index;

[0018] The acoustic data in the first acoustic information that matches the collection time of the second fish image is extracted as the second acoustic information, and the time deviation of the matching time is not more than the preset synchronization threshold;

[0019] The environmental monitoring data corresponding to the collection time of the second fish image is selected from the first environmental parameter as the second environmental parameter;

[0020] The time sequence correlation between the second fish image, the second acoustic information and the second environmental parameter is established, so that they have a unified time coding identifier;

[0021] The second fish image is subjected to optical compensation processing to eliminate image quality degradation caused by water scattering;

[0022] The second acoustic information is subjected to noise suppression processing to retain the fish biological characteristic signals of the effective frequency band;

[0023] The processed fish information is generated, and the plurality of processed fish information is stored in the preprocessing database.

[0024] In some embodiments, a three-dimensional point cloud reconstruction is performed on the second fish image, and static phenotype parameters are extracted, including fish body surface scale arrangement features and fin strip morphology features, including:

[0025] A multi-view fish body surface image is selected from the second fish image as input data for three-dimensional reconstruction, and dense point cloud data of the fish body surface is generated, which contains three-dimensional spatial coordinates and texture information;

[0026] Fish body contour feature points are extracted from the dense point cloud data, and a three-dimensional mesh model of the fish body is constructed;

[0027] The scale distribution area is calibrated on the three-dimensional mesh model of the fish body, and the scale arrangement features are extracted, including scale spacing and scale arrangement density;

[0028] The fin strip structure in the three-dimensional mesh model is identified, and the fin strip morphology features are extracted, including the size information of each fin strip, including length, angle and relative position;

[0029] The scale arrangement features and fin strip morphology features are associated with the fish individual identification information and stored to form structured phenotype features;

[0030] In addition, surface curvature analysis is performed on the three-dimensional mesh model, and morphological parameters of each part of the fish body are extracted as supplementary phenotype features;

[0031] The supplementary phenotype features and the structured phenotype features are sorted into static phenotype parameters.

[0032] In some embodiments, multispectral analysis is performed on the second fish image to obtain body surface texture reflection characteristics, including:

[0033] Fish body surface image data of different spectral bands is separated from the second fish image to obtain a multispectral image data set, and the spectral bands include visible light bands and near-infrared bands;

[0034] The multispectral image data set is normalized to obtain a standardized multispectral image, which eliminates the influence of light intensity difference on reflection characteristics;

[0035] Reflectivity data of the fish body back, lateral line and abdomen are extracted from the standardized multispectral image to generate reflectivity curves of each part;

[0036] Based on the reflectivity curves, reflectivity feature maps of different parts of the fish body are constructed, which contain the corresponding relationship between wavelength and reflectivity intensity;

[0037] Texture features are extracted from the standardized multispectral image, and gray level co-occurrence matrix features and local binary pattern features under each spectral band are calculated to obtain multispectral texture feature parameters;

[0038] The reflectance feature spectrum and the multi-spectral texture feature parameter are stored in association with the fish individual identification information to form a structured texture feature;

[0039] The multi-spectral texture feature parameter is fused to extract a comprehensive texture feature across bands as a supplementary texture feature;

[0040] The supplementary texture feature and the structured texture feature are arranged into a body surface texture reflection characteristic.

[0041] In some embodiments, the second fish image and the second acoustic signal are correlated in time domain to extract a dynamic behavior feature, and a behavior feature encoding library containing a typical behavior mode is established, including:

[0042] The second fish image sequence is analyzed for inter-frame motion to obtain fish body motion trajectory data, the fish body motion trajectory data containing position, velocity and acceleration parameters;

[0043] The second acoustic information is analyzed for time-frequency to extract a swim bladder vibration feature parameter, the swim bladder vibration feature parameter including vibration frequency, vibration amplitude and vibration duration;

[0044] The fish body motion trajectory data and the swim bladder vibration feature parameter are time-synchronously matched to establish a motion-acoustic joint feature matrix;

[0045] A typical behavior mode feature is identified from the motion-acoustic joint feature matrix, the typical behavior mode including feeding behavior feature, escape behavior feature and fighting behavior feature;

[0046] The typical behavior mode feature is parameterized and encoded to generate a behavior feature encoding vector, the encoding vector containing combined parameters of motion trajectory feature and acoustic vibration feature;

[0047] The behavior feature encoding vector is stored in association with the fish individual identification information to form a structured behavior feature library;

[0048] The behavior feature encoding vector is clustered and analyzed to establish a behavior mode classification model as a supplementary behavior feature;

[0049] The supplementary behavior feature and the structured behavior feature library are arranged into a behavior feature encoding library.

[0050] In some embodiments, a mapping relationship network of different disease types and body surface lesion area spatial distribution, behavior abnormal mode is constructed according to the static phenotype parameter, the body surface texture reflection characteristic, the dynamic behavior feature and the behavior feature encoding library, including:

[0051] A lesion-related morphological feature is extracted from the static phenotype parameter, the lesion-related morphological feature including scale abnormal shedding area and fin strip damage degree;

[0052] analyzing the abnormal reflection area in the reflection characteristic of the body surface texture, identifying the spatial distribution characteristic of the body surface lesion, the spatial distribution characteristic including a lesion area proportion and a lesion area geometric characteristic;

[0053] screening an abnormal behavior parameter from the dynamic behavior characteristic, the abnormal behavior parameter including a feeding frequency drop amplitude and a swimming trajectory disorder degree;

[0054] correlating the lesion related morphological characteristic, the body surface lesion spatial distribution characteristic and the behavior abnormal parameter, and establishing a disease-phenotype-behavior three-dimensional characteristic space;

[0055] constructing a disease type discrimination model in the three-dimensional characteristic space, and the discrimination model outputting a mapping relationship between the disease type and the characteristic combination;

[0056] correlating the disease type discrimination model with a behavior characteristic code library, and generating an abnormal behavior characteristic set containing a typical disease behavior mode;

[0057] and, performing visual processing on the abnormal behavior characteristic set, and constructing a disease characteristic map as a supplementary diagnosis basis;

[0058] integrating the disease characteristic map, the disease type discrimination model and the abnormal behavior characteristic set to form a disease characteristic mapping relationship network.

[0059] In some embodiments, based on the static phenotype parameter and the dynamic behavior characteristic, establishing a characteristic parameter conversion relationship between the reference fish and the target fish includes:

[0060] extracting morphological characteristic principal components of the reference fish and the target fish from the static phenotype parameter by using a principal component analysis method, the morphological characteristic principal components including a scale arrangement characteristic vector and a fin strip morphological characteristic vector;

[0061] extracting behavior characteristic sequences of the reference fish and the target fish from the dynamic behavior characteristic by using a dynamic time warping algorithm, the behavior characteristic sequences including a standardized movement trajectory and a normalized swim bladder vibration spectrum;

[0062] calculating a conversion matrix between the morphological characteristic principal components of the reference fish and the target fish by using a canonical correlation analysis method, and obtaining a morphological characteristic conversion relationship;

[0063] applying the dynamic time warping algorithm to calculate an alignment path between the behavior characteristic sequences of the reference fish and the target fish, and obtaining a behavior characteristic conversion relationship;

[0064] based on the morphological characteristic conversion relationship and the behavior characteristic conversion relationship, constructing the characteristic parameter conversion relationship from the reference fish to the target fish.

[0065] In some embodiments, the static phenotype parameters, dynamic behavior characteristics, behavior characteristic code library, mapping relationship network, and characteristic parameter conversion relationship are stored in a structured manner, and a multi-dimensional feature database containing fish phenotype characteristics, behavior characteristics, disease characteristics, and environmental parameters is constructed, including:

[0066] A storage architecture is established based on NoSQL, containing a three-level index structure of fish individual identification, collection timestamp, and environmental parameters;

[0067] The scale arrangement characteristics, fin strip morphology characteristics, and supplementary phenotype characteristics in the static phenotype parameters are stored in association with the corresponding three-dimensional grid models;

[0068] The motion trajectory characteristics, swim bladder vibration characteristics, and behavior pattern classification in the dynamic behavior characteristics are stored in time sequence coding;

[0069] The typical behavior pattern characteristics in the behavior characteristic code library are converted into vector data, and a behavior characteristic association network is constructed;

[0070] The disease types, body surface lesion characteristics, and behavior abnormal patterns in the mapping relationship network are stored in association;

[0071] The characteristic parameter conversion relationship is stored in matrix form, and a mapping relationship is established with the fish species classification label;

[0072] The various types of feature data in the storage architecture are integrated to generate a multi-dimensional feature database.

[0073] The multi-dimensional feature database realizes cross-fish species application through the characteristic parameter conversion relationship, for example, migrating the disease-behavior mapping model of large yellow croaker to grouper or grouper, which significantly improves the disease early warning efficiency in net cage culture scenarios. The database supports breeding modes including intensive pond, net cage, and roadbed culture, and is especially suitable for precise health management of high-value fish.

[0074] In a second aspect, the present application also provides a live fish underwater feature database system, which is suitable for the method of the first aspect, and the system comprises:

[0075] An information collection module is used to collect original fish information at a preset frequency, and the original fish information includes first fish images, first acoustic information, and first environmental parameters, and the original fish information is preprocessed to obtain processed fish information, and the preprocessing includes data cleaning and spatio-temporal alignment, and the processed fish information includes second fish images, second acoustic information, and second environmental parameters after spatio-temporal alignment;

[0076] The feature extraction module is configured to perform three-dimensional point cloud reconstruction on the second fish image, extract static phenotype parameters, and perform multispectral analysis on the second fish image to obtain body surface texture reflection characteristics, wherein the static phenotype parameters include fish body surface scale arrangement characteristics and fin strip morphology characteristics; the second fish image and the second acoustic signal are correlated in time domain to extract dynamic behavior characteristics, and a behavior characteristic code library containing typical behavior modes is established, wherein the dynamic behavior characteristics include fish body movement trajectory and swim bladder vibration frequency, and the typical behavior modes include at least one of feeding, escaping and fighting;

[0077] The database construction module is configured to construct a mapping relationship network of different disease types and body surface lesion region spatial distribution and behavior abnormal mode according to the static phenotype parameters, the body surface texture reflection characteristics, the dynamic behavior characteristics and the behavior characteristic code library; to establish a feature parameter conversion relationship between the reference fish species and the target fish species based on the static phenotype parameters and the dynamic behavior characteristics; and to store the static phenotype parameters, the dynamic behavior characteristics, the behavior characteristic code library, the mapping relationship network and the feature parameter conversion relationship in a structured manner, and to construct a multi-dimensional feature database containing fish phenotype characteristics, behavior characteristics, disease characteristics and environmental parameters.

[0078] In a third aspect, the present application further provides a computer readable storage medium, which stores computer program instructions, and the computer program instructions, when executed by a processor, implement the method of the first aspect.

[0079] Distinguishing from the prior art, the technical scheme provides a method, system and medium for constructing an underwater feature database of live fish. The method collects original fish information through a preset frequency, extracts static phenotype parameters such as scale arrangement features and fin strip morphological features of a fish body surface from a second fish image obtained after preprocessing through three-dimensional point cloud reconstruction, and obtains body surface texture reflection characteristics through multispectral analysis. Meanwhile, the method extracts dynamic behavior features such as movement trajectories and swim bladder vibration frequencies by associating time domain features of the second fish image and the second acoustic signal, establishes a feature coding library containing typical behavior patterns, constructs a mapping relationship network of different disease types and body surface lesion area spatial distribution and behavior abnormal patterns based on the above features, and establishes a feature parameter conversion relationship. Finally, the method structurally stores various feature parameters to form a multi-dimensional feature database containing fish phenotype features, behavior features, disease features and environmental parameters, realizes multi-dimensional correlation analysis of fish static phenotypes and dynamic behaviors, and improves the precision of disease diagnosis, health breeding and cross-species feature migration. The multi-dimensional feature database constructed by the method can be widely applied to health monitoring of economic fish such as large yellow croaker (Larimichthys crocea), kelp bass (Lateolabrax japonicus), grouper (Epinephelus spp.), eel (Anguilla spp.) and sturgeon (Acipenser sturio Linnaeus), and is suitable for various breeding modes such as intensive culture ponds, net cage culture and water-based roadbed culture, thereby providing standardized data support for disease diagnosis and behavior analysis in different breeding environments.

[0080] The above invention content is only a summary of the technical scheme of the present application. In order to enable those skilled in the art to more clearly understand the technical scheme of the present application, and then implement the content recorded in the specification and drawings, and in order to enable the above and other purposes, features and advantages of the present application to be more easily understood, the following will be described in combination with the specific embodiments of the present application and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0081] The drawings are only used to show the principles, implementation modes, applications, characteristics and effects of the specific embodiments and other related contents of the present application, and cannot be considered as limitations of the present application.

[0082] In the drawings of the specification:

[0083] Figure 1 A method step diagram of steps S101 to S107 of the method for constructing the underwater feature database described in the specific embodiments;

[0084] Figure 2 A structure schematic diagram of the underwater feature database system described in the specific embodiments.

[0085] The reference signs mentioned in the above-mentioned figures are explained as follows:

[0086] 1. An underwater feature database system;

[0087] 11. An information collection module;

[0088] 12. A feature extraction module;

[0089] 13. A database construction module. DETAILED DESCRIPTION

[0090] To make the possible application scenarios, technical principles, specific schemes that can be implemented, purposes and effects achieved, etc. of the present application clear, the following will be described in detail in combination with the specific embodiments listed and the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical schemes of the present application, and therefore only serve as examples, but cannot limit the protection scope of the present application.

[0091] In this article, the term "embodiment" means that the specific features, structures or characteristics described in combination with the embodiment can be included in at least one embodiment of the present application. The term "embodiment" appearing at various positions in the specification does not necessarily refer to the same embodiment, and does not particularly limit the independence or association between other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the technical features mentioned in each embodiment can be combined in any way to form a corresponding implementable technical scheme.

[0092] Unless otherwise defined, the meanings of the technical terms used herein are the same as those commonly understood by those skilled in the art to which the present application belongs; the use of related terms herein is only for the purpose of describing specific embodiments, and is not intended to limit the present application.

[0093] In the description of the present application, the phrase "and / or" is a description of the logical relationship between objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this article generally represents that the associated objects before and after are a "or" logical relationship.

[0094] In the present application, the phrases such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, primary and secondary or order relationship between the entities or operations.

[0095] In the absence of more limitations, in this application, the "include", "contain", "have" or other similar open expressions used in the statements are intended to cover the non-exclusive inclusion, and these expressions do not exclude the presence of other elements in the process, method or product including the elements, so that the process, method or product including a series of elements can not only include those limited elements, but also include other elements not explicitly listed, or also include the elements inherent in such process, method or product.

[0096] As the same understanding in the "Examination Guidelines", in this application, the expressions such as "greater than", "less than", "exceed" are understood as not including the number; the expressions such as "above", "below", "within" are understood as including the number. In addition, in the description of the embodiments of the application, the meaning of "multiple" is more than two (including two), and similar expressions related to "multiple" are also understood in this way, for example, "multiple groups", "multiple times" and the like, unless otherwise explicitly limited.

[0097] In the description of the embodiments of the application, the spatial-related expressions used, such as "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", and the like, indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or the drawings, and are only for the convenience of describing the specific embodiments of the application or for the reader to understand, and do not indicate or imply that the indicated device or component must have a particular position, a particular orientation, or be constructed or operated in a particular orientation, therefore cannot be understood as a limitation of the embodiments of the application.

[0098] The processor described in the embodiments of the present application can be implemented by hardware, firmware, software or a combination thereof, and can use at least one of circuit, single or multiple application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), central processing units (CPU), controllers, microcontrollers, microprocessors, and other physical, biological or chemical structures that can realize the same or equivalent functions as the above-mentioned processors, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute part or all steps or any combination of steps mentioned in the computer programs or methods of various embodiments of the present application.

[0099] Please refer to Figure 1 In the first aspect, the embodiments of the present application provide a method for constructing an underwater feature database of live fish, comprising:

[0100] S101, collecting original fish information according to a preset frequency, the original fish information including a first fish image, first acoustic information and first environmental parameters;

[0101] S102, preprocessing the original fish information to obtain processed fish information, the preprocessing including data cleaning and spatio-temporal alignment, the processed fish information including a second fish image, second acoustic information and second environmental parameters after spatio-temporal alignment;

[0102] S103, performing three-dimensional point cloud reconstruction on the second fish image to extract static phenotypic parameters, the static phenotypic parameters including fish body surface scale arrangement features and fin strip morphological features, and performing multispectral analysis on the second fish image to obtain body surface texture reflection characteristics;

[0103] S104, performing time domain correlation on the second fish image and the second acoustic signal to extract dynamic behavior features, the dynamic behavior features including fish body movement trajectory and swim bladder vibration frequency, and establishing a behavior feature code library containing typical behavior patterns, the typical behavior patterns including at least one of feeding, escaping and fighting;

[0104] S105. Construct a mapping relationship network of different disease types and spatial distribution of body surface lesion area and behavior abnormality mode according to the static phenotype parameters, body surface texture reflection characteristics, dynamic behavior characteristics, and behavior characteristic code library.

[0105] S106. Establish a characteristic parameter conversion relationship between the reference fish and the target fish based on the static phenotype parameters and the dynamic behavior characteristics.

[0106] S107. Structurally store the static phenotype parameters, dynamic behavior characteristics, behavior characteristic code library, mapping relationship network, and characteristic parameter conversion relationship, and construct a multi-dimensional characteristic database containing fish phenotype characteristics, behavior characteristics, disease characteristics, and environmental parameters.

[0107] In step S101, the preset frequency is a data acquisition time interval preset according to the fish activity law and monitoring demand, the first fish image is original image data obtained by an underwater optical imaging device, and the first acoustic information is fish sound and swimming signals collected by a hydroacoustic detection device. Preferably, the first environmental parameters include water temperature, dissolved oxygen, pH value, and other water physical and chemical indicators.

[0108] In step S102, invalid data such as blur and occlusion is removed through data cleaning. The spatio-temporal alignment is to establish the spatio-temporal correlation of the first fish image, the first acoustic information, and the first environmental parameters through time stamp synchronization and spatial coordinate registration, so as to obtain the standardized data after alignment, i.e., the second fish image, the second acoustic information, and the second environmental parameters.

[0109] In step S103, the three-dimensional point cloud reconstruction is to reconstruct the three-dimensional morphology of the fish body through multi-view image matching and three-dimensional modeling technology, the fish body surface scale arrangement characteristics refer to geometric characteristics such as scale spacing and arrangement law, the fin strip morphology characteristics include fin strip length and curvature, and the body surface texture reflection characteristics are different waveband light reflection characteristics obtained through multi-spectral imaging analysis.

[0110] In step S104, the second fish image and the second acoustic signal are time-domain correlated, and the second fish image and the second acoustic signal are matched and analyzed according to the time sequence. Preferably, the fish body movement trajectory is obtained by continuous frame image tracking, the swim bladder vibration frequency is determined by acoustic signal spectrum analysis, and the behavior characteristic code library is a database in which behavior patterns such as feeding, escaping, and fighting are feature-coded and classified stored through machine learning.

[0111] In step S105, the mapping relationship network is a correlation model of disease types and body surface lesion area (such as ulcer site) and behavior abnormality mode (such as slow swimming) established through deep learning.

[0112] In step S106, a mathematical conversion model between parameters, i.e., a characteristic parameter conversion relationship, is established to realize cross-species migration application of the characteristic parameters, thereby expanding the application range of the database.

[0113] In step S107, each type of feature is classified and stored according to a preset data architecture, so as to ensure ordered organization and efficient management of the multi-source heterogeneous data.

[0114] The implementation principle of the embodiment can be understood as follows: a digital representation system of fish characteristics is established through multi-source data acquisition and standardized processing. Specifically, the method first synchronously acquires optical, acoustic and environmental monitoring data to ensure that the first fish image, the first acoustic information and the first environmental parameter have spatiotemporal consistency; then, the second fish image is converted into quantifiable static phenotype parameters by using three-dimensional reconstruction technology, and subtle surface texture features are captured through multi-spectral analysis; at the behavior analysis level, the second fish image and the second acoustic signal are time-matched to extract dynamic behavior features with diagnostic value and establish standardized coding. These feature data are finally integrated into an intelligent diagnostic model with clear pathological correlation, establishing an interpretable correlation between the apparent characteristics and the pathological state. For example, when a specific abnormal scale arrangement feature is detected accompanied by a disorder of the swim bladder vibration frequency, the possible disease type can be quickly located in combination with the mapping relationship network, realizing a conversion closed loop from raw data to diagnostic knowledge.

[0115] The embodiment ensures the spatiotemporal consistency of multi-source data by synchronously acquiring the first fish image, the first acoustic information and the first environmental parameter and implementing spatiotemporal alignment processing; realizes comprehensive digital representation of fish phenotype characteristics by accurately extracting static phenotype parameters such as scale arrangement features and fin strip morphological features of the fish body surface, and combining the surface texture reflection characteristics obtained through multi-spectral analysis; provides a standardized basis for fish behavior analysis by obtaining dynamic behavior features such as fish body movement trajectory and swim bladder vibration frequency through time-domain correlation analysis of the second fish image and the second acoustic signal, and combining the established behavior feature coding library; significantly improves the accuracy and efficiency of disease diagnosis by constructing a mapping relationship network of disease types and spatial distribution of body surface lesions, abnormal behavior patterns; expands the application range of the database by establishing a conversion relationship of characteristic parameters between the reference fish species and the target fish species; finally, realizes ordered integration and efficient management of fish phenotype characteristics, behavior characteristics, disease characteristics and environmental parameters through the structured storage of the multi-dimensional feature database, thereby providing reliable data support for fish health monitoring and disease diagnosis.

[0116] In some embodiments, the original fish information is preprocessed to obtain processed fish information, the preprocessing including data cleaning and spatiotemporal alignment, and the processed fish information including the second fish image, the second acoustic information and the second environmental parameter after spatiotemporal alignment.

[0117] The first fish image that meets the preset definition threshold is selected as the second fish image, and the preset definition threshold includes edge sharpness and contrast index;

[0118] The acoustic data in the first acoustic information that matches the second fish image acquisition time is extracted as the second acoustic information, and the time deviation in the matching time is not more than the preset synchronization threshold;

[0119] The environmental monitoring data corresponding to the second fish image acquisition time is selected from the first environmental parameter as the second environmental parameter;

[0120] The time sequence correlation between the second fish image, the second acoustic information and the second environmental parameter is established, so that they have a unified time coding identifier;

[0121] The second fish image is optically compensated to eliminate image quality degradation caused by water scattering;

[0122] The second acoustic information is subjected to noise suppression processing to retain the fish biological feature signal of the effective frequency band;

[0123] The processed fish information is generated, and the plurality of processed fish information is stored in the preprocessing database.

[0124] In this embodiment, the preset definition threshold is a quantitative standard for screening the first fish image, and the usability of the image is represented by edge sharpness and contrast index, wherein the edge sharpness is used to evaluate the sharpness of the fish body contour, and the contrast index is used to judge the distinction degree of the target and the background in the image.

[0125] The preset synchronization threshold refers to the upper limit of the time deviation allowed between the first acoustic information and the second fish image acquisition time, preferably, the preset synchronization threshold is determined according to the response time of the data acquisition device, the signal transmission delay and the fish behavior characteristic change rate, and is usually set to not more than 1 / 3 of the interval between adjacent two frames of first fish image acquisition, to ensure that the time deviation of acoustic and optical data does not affect the accuracy of behavior characteristic analysis.

[0126] The establishment of the time sequence correlation is realized by giving the second fish image, the second acoustic information and the second environmental parameter a unified time coding identifier, and this correlation ensures the strict correspondence of multi-source data in the time dimension.

[0127] The generation process of the processed fish information is essentially a process of systematically integrating various types of data after screening, matching, compensation and inhibition processing. The optical compensation processing is an underwater imaging quality optimization method implemented for the second fish image. The optical distortion and contrast reduction caused by water scattering are eliminated through a compensation algorithm. The noise suppression processing refers to the signal enhancement operation on the second acoustic information. The environmental noise interference is removed through filtering technology, and the effective frequency band signal reflecting the biological characteristics of fish is retained. The preprocessing database is a structured storage system specially used for storing the standardized processed fish information.

[0128] The present embodiment converts the original, heterogeneous first fish image, first acoustic information and first environmental parameter into standardized processed fish information with strict spatiotemporal correlation through multi-level preprocessing operations. For example, when the edge sharpness of a certain frame of the first fish image is detected to be lower than the preset clarity threshold, the frame of image is automatically excluded to ensure the image quality of the input subsequent analysis. This refined preprocessing process lays a reliable data foundation for subsequent feature extraction and analysis.

[0129] The present embodiment ensures the reliability of image analysis by screening the second fish image through the preset clarity threshold, ensures the temporal consistency of acoustic and optical data by matching the second acoustic information with the preset synchronization threshold, establishes the temporal correlation relationship to realize the multi-source data fusion of the second fish image, the second acoustic information and the second environmental parameter. The optical compensation processing improves the quality of the second fish image, and the noise suppression processing optimizes the signal-to-noise ratio of the second acoustic information. The finally generated processed fish information is standardized stored through the preprocessing database, providing a high-quality data foundation for subsequent feature extraction. The preprocessing process of the present embodiment effectively solves the heterogeneity problem of underwater environmental data acquisition, and significantly improves the accuracy and reliability of fish feature analysis.

[0130] In some embodiments, the second fish image is subjected to three-dimensional point cloud reconstruction, and static phenotype parameters are extracted, including fish body surface scale arrangement features and fin strip morphological features, including:

[0131] A multi-view fish body surface image is selected from the second fish image as input data for three-dimensional reconstruction, generating dense point cloud data of the fish body surface, which contains three-dimensional spatial coordinates and texture information;

[0132] Fish body contour feature points are extracted from the dense point cloud data, and a three-dimensional mesh model of the fish body is constructed;

[0133] The scale distribution area is calibrated on the three-dimensional mesh model of the fish body, and the scale arrangement features are extracted, including scale spacing and scale arrangement density;

[0134] Identify the fin structure in the three-dimensional mesh model, extract the fin shape feature, the fin shape feature includes the size information of each fin, the size information of the fin includes the length, angle and relative position;

[0135] Store the scale arrangement feature and the fin shape feature in association with the fish individual identification information to form a structured phenotype feature;

[0136] And, perform surface curvature analysis on the three-dimensional mesh model to extract the morphological parameters of each part of the fish body as a supplementary phenotype feature;

[0137] Organize the supplementary phenotype feature and the structured phenotype feature into static phenotype parameters.

[0138] In the embodiment, the three-dimensional reconstruction input data refers to the image sequence selected from the second fish image in accordance with the multi-view overlapping requirement, preferably, the images are generated into initial point cloud through SFM (structure from motion) algorithm, and then optimized into dense point cloud data through MVS (multi-view stereo matching) algorithm. Further, the fish three-dimensional mesh model can be converted into continuous surface through surface reconstruction algorithm such as Poisson reconstruction, and the model completely retains the three-dimensional space coordinates and texture information of the original point cloud.

[0139] Preferably, the extraction of the scale arrangement feature adopts the region growing algorithm based on the three-dimensional mesh model, first identifies the scale boundary through curvature analysis, then calculates the distance between the mass centers of adjacent scales as the scale distance, and counts the number of scales per unit area to obtain the arrangement density. Preferably, the extraction of the fin shape feature combines three-dimensional edge detection and skeleton extraction algorithm, first locates the fin region, and then calculates the geometric parameters thereof. The above feature extraction algorithms are all based on a unified three-dimensional mesh model, ensuring the consistency of the data source.

[0140] Further, the surface curvature analysis can adopt the moving least squares method to calculate the curvature of the mesh vertex, and the obtained morphological parameters have the same geometric reference as the aforementioned features. All the feature parameters are associated with the fish individual identification information through database indexing to form a hierarchical static phenotype parameter system.

[0141] The embodiment is closely connected with the aforementioned preprocessing link, the quality of the second fish image directly affects the three-dimensional reconstruction accuracy, and the generated static phenotype parameters will serve the subsequent dynamic behavior analysis and disease diagnosis. For example, when the system processes the preprocessed second fish image, a high-precision three-dimensional model is obtained through multi-view reconstruction, scale and fin features are extracted based on the model, and finally various parameters are integrated and stored, ensuring the data consistency from image acquisition to feature extraction. The three-dimensional reconstruction link provides a unified data basis for subsequent analysis, and the feature extraction algorithm is specially optimized for different phenotype features, ensuring the accuracy of feature extraction while maintaining the coordination of the method.

[0142] In some embodiments, the multispectral analysis of the second fish image to obtain the body surface texture reflection characteristics comprises:

[0143] Separating fish body surface image data of different spectral bands from the second fish image to obtain a multispectral image data set, the spectral bands including a visible light band and a near-infrared band;

[0144] Normalizing the multispectral image data set to obtain a standardized multispectral image, eliminating the influence of light intensity difference on the reflection characteristics;

[0145] Extracting reflectance data of the fish body back, lateral line and abdomen from the standardized multispectral image to generate reflectance curves of each part;

[0146] Constructing reflectance feature maps of different parts of the fish body based on the reflectance curves, the reflectance feature maps containing the correspondence between wavelength and reflection intensity;

[0147] Extracting texture features from the standardized multispectral image, calculating the gray level co-occurrence matrix features and local binary pattern features under each spectral band to obtain multispectral texture feature parameters;

[0148] Storing the reflectance feature maps and the multispectral texture feature parameters in association with the fish individual identification information to form structured texture features;

[0149] And, performing feature fusion on the multispectral texture feature parameters to extract comprehensive texture features across bands as supplementary texture features;

[0150] Organizing the supplementary texture features and the structured texture features into the body surface texture reflection characteristics.

[0151] In the present embodiment, the multispectral image data set can be separated by optical filtering technology, in which the visible light band is used to capture regular color information and the near-infrared band is used to obtain deep tissue features.

[0152] Preferably, the standardized multispectral image is the image data after eliminating environmental light differences by a radiation correction algorithm, ensuring the accuracy of reflectance measurement; the reflectance curve is the reflectance intensity distribution of a specific part of the fish body at different wavelengths extracted by a spectral analysis algorithm, representing the optical properties of the part.

[0153] The reflectance feature map is a two-dimensional chart generated by spatial mapping of the reflectance curves of different parts of the fish body, in which the horizontal axis represents the wavelength and the vertical axis represents the reflection intensity, for intuitively displaying the spectral reflection characteristic differences of different parts.

[0154] The gray level co-occurrence matrix feature is a texture description parameter obtained by counting the spatial relationship of image pixels, the local binary pattern feature is a local texture pattern extracted by comparing the intensity relationship of pixel neighborhood, and the multispectral texture feature parameter refers to a set of the above-mentioned texture features extracted from each waveband image.

[0155] Preferably, the feature fusion of the multispectral texture feature parameter can integrate the multi-waveband texture parameters into more representative comprehensive features through a dimension reduction algorithm such as principal component analysis.

[0156] The present embodiment can obtain optical fingerprint information of the fish body surface through multispectral imaging technology, which can be understood in combination with the following examples: when processing a carp image, first separate the feature waveband images of 450 nm, 550 nm, 650 nm, etc., extract the reflectivity curve of the back region after normalization, and then combine the texture features of each waveband to finally form the complete surface texture reflection characteristics, so as to comprehensively represent the biologic optical characteristics of the fish body surface.

[0157] The present embodiment extracts accurate surface texture reflection characteristics from the second fish image through multispectral analysis technology, the standardized multispectral image ensures the comparability of the reflectivity data, the reflectivity feature spectrum intuitively presents the optical characteristic differences of different parts of the fish body, the multispectral texture feature parameter comprehensively describes the texture features of each waveband, and the supplementary texture feature generated by feature fusion improves the representation ability of the feature. The present embodiment realizes multi-scale and multi-dimensional representation of the optical characteristics of the fish body surface, and provides reliable optical feature basis for fish species identification and health status evaluation.

[0158] In some embodiments, the second fish image and the second acoustic signal are correlated in time domain to extract dynamic behavior features, and a behavior feature code library containing typical behavior modes is established, including:

[0159] The second fish image sequence is analyzed for inter-frame motion to obtain fish body motion trajectory data, and the fish body motion trajectory data contains position, velocity and acceleration parameters;

[0160] The second acoustic information is analyzed for time-frequency to extract swim bladder vibration feature parameters, and the swim bladder vibration feature parameters include vibration frequency, vibration amplitude and vibration duration;

[0161] The fish body motion trajectory data and the swim bladder vibration feature parameters are time-synchronously matched to establish a motion-acoustic joint feature matrix;

[0162] Typical behavior mode features are identified from the motion-acoustic joint feature matrix, and the typical behavior modes include feeding behavior features, escape behavior features and fighting behavior features;

[0163] The typical behavior pattern features are parameterized and coded to generate behavior feature coding vectors, which include combined parameters of motion trajectory features and acoustic vibration features;

[0164] The behavior feature coding vectors are stored in association with the fish individual identification information to form a structured behavior feature library;

[0165] In addition, the behavior feature coding vectors are subjected to cluster analysis to establish a behavior pattern classification model as a supplementary behavior feature;

[0166] The supplementary behavior feature and the structured behavior feature library are sorted into a behavior feature coding library.

[0167] In the embodiment, the inter-frame motion analysis can calculate the continuous displacement of the fish body in the second fish image sequence by an optical flow method or a feature point tracking algorithm, so as to obtain fish body motion trajectory data including position, velocity and acceleration parameters.

[0168] The time-frequency analysis can process the second acoustic signal by a short-time Fourier transform algorithm or the like to extract parameters of the swim bladder vibration features reflecting the physiological state of the fish, wherein the vibration frequency represents the sound emission characteristics, the vibration amplitude reflects the sound emission intensity, and the vibration duration indicates the behavior duration.

[0169] Further, the fish body motion trajectory data and the swim bladder vibration feature parameters are integrated to form a multi-dimensional feature set, i.e., a motion-acoustic joint feature matrix, by time stamp alignment.

[0170] The recognition of the typical behavior pattern features is a pattern recognition process based on the motion-acoustic joint feature matrix, the feeding behavior features are represented by specific motion trajectory and swim bladder vibration combinations, and the escape behavior features and the fighting behavior features each have a unique motion-acoustic feature combination.

[0171] Preferably, the typical behavior pattern features are converted into a standardized parameter set, i.e., a behavior feature coding vector, by a dimension reduction algorithm such as principal component analysis, and the behavior pattern classification model is a classification system established by cluster analysis of the behavior feature coding vectors.

[0172] The embodiment combines the second fish image and the second acoustic signal by time domain correlation analysis, extracts fish body motion trajectory data and swim bladder vibration feature parameters that comprehensively represent the dynamic behavior features of the fish, establishes a motion-acoustic joint feature matrix to realize fusion analysis of multi-modal behavior features, generates a behavior feature coding vector and a structured behavior feature library to provide a standardized basis for fish behavior recognition, and further improves the accuracy of behavior analysis by establishing a behavior pattern classification model as a supplementary behavior feature.

[0173] In some embodiments, constructing the mapping relationship network of different disease types and spatial distribution of body surface lesion area, abnormal behavior pattern according to the static phenotype parameters, body surface texture reflection characteristics, dynamic behavior characteristics and behavior characteristic encoding library comprises:

[0174] extracting lesion-related morphological features from static phenotype parameters, the lesion-related morphological features including scale abnormal shedding area and fin damage degree;

[0175] analyzing abnormal reflection areas in body surface texture reflection characteristics to identify spatial distribution characteristics of body surface lesions, the spatial distribution characteristics including lesion area proportion and lesion area geometric characteristics;

[0176] screening abnormal behavior parameters from dynamic behavior characteristics, the abnormal behavior parameters including feeding frequency decrease amplitude and swimming trajectory disorder degree;

[0177] correlating and analyzing the lesion-related morphological features, the spatial distribution characteristics of body surface lesions and the abnormal behavior parameters to establish a disease-phenotype-behavior three-dimensional characteristic space;

[0178] constructing a disease type discrimination model in the three-dimensional characteristic space, the discrimination model outputting a mapping relationship of disease type and characteristic combination;

[0179] correlating the disease type discrimination model with the behavior characteristic encoding library to generate an abnormal behavior characteristic set containing typical disease behavior patterns;

[0180] and, visualizing the abnormal behavior characteristic set to construct a disease characteristic map as a supplementary diagnosis basis;

[0181] integrating the disease characteristic map, the disease type discrimination model and the abnormal behavior characteristic set to form a disease characteristic mapping relationship network.

[0182] In this embodiment, the lesion-related morphological features are morphological abnormalities related to diseases identified through static phenotype parameter analysis, wherein the scale abnormal shedding area is determined by scale loss detection on the three-dimensional grid model, and the fin damage degree is calculated according to the deviation of fin morphological characteristics from the reference value.

[0183] The spatial distribution characteristics of body surface lesions are quantitative parameters obtained by analyzing abnormal reflection areas in body surface texture reflection characteristics, the lesion area proportion is calculated by an image segmentation algorithm to obtain the proportion of abnormal areas in the total body surface area, and the lesion area geometric characteristics include parameters such as shape complexity of lesion boundary and spatial distribution regularity.

[0184] The abnormal behavior parameter is an index deviating from a normal range selected from the dynamic behavior feature, wherein a feeding frequency reduction amplitude is derived by comparing a current feeding behavior feature with a reference value in the behavior feature code library, and a swimming trajectory disorder degree is quantified by analyzing regularity changes of the fish body movement trajectory data.

[0185] Further, the lesion-related morphological features, the body surface lesion spatial distribution features, and the abnormal behavior parameter are integrated by multivariate statistical analysis to form a disease-phenotype-behavior three-dimensional feature space.

[0186] Preferably, a classifier, i.e., a disease type discrimination model, is established in the three-dimensional feature space by a machine learning algorithm, which can output a corresponding disease type prediction according to an input feature combination.

[0187] The disease feature atlas is a diagnostic auxiliary tool for graphically expressing the relationship between the abnormal behavior feature set and the disease feature by data visualization technology; and the disease feature mapping relationship network is a comprehensive diagnostic knowledge system integrating the disease feature atlas, the disease type discrimination model, and the abnormal behavior feature set.

[0188] The embodiment integrates the static phenotype parameter, the body surface texture reflection characteristic, and the dynamic behavior feature to construct a mapping relationship network fully reflecting the disease feature. The lesion-related morphological features and the body surface lesion spatial distribution features provide intuitive pathological characterization, the abnormal behavior parameter realizes the behavior quantitative of the disease, the establishment of the three-dimensional feature space realizes the organic integration of the multi-source features, the disease type discrimination model provides accurate disease diagnosis capability, and the abnormal behavior feature set and the disease feature atlas enhance the explainability of the diagnosis. The embodiment realizes the intelligent mapping from the appearance feature to the disease diagnosis, and provides a scientific basis for fish health monitoring.

[0189] In some embodiments, based on the static phenotype parameter and the dynamic behavior feature, the establishment of the feature parameter conversion relationship between the reference fish and the target fish includes:

[0190] The principal component analysis method is used to extract morphological feature principal components of the reference fish and the target fish from the static phenotype parameter, and the morphological feature principal components include scale arrangement characteristic vectors and fin strip morphological characteristic vectors;

[0191] The dynamic behavior feature sequence of the reference fish and the target fish is extracted from the dynamic behavior feature by the dynamic time warping algorithm, and the behavior feature sequence includes standardized movement trajectories and normalized swim bladder vibration spectra;

[0192] The conversion matrix between the morphological feature principal components of the reference fish and the target fish is calculated by using the canonical correlation analysis method, and the morphological feature conversion relationship is obtained;

[0193] The dynamic time warping algorithm is applied to calculate an alignment path between the behavior characteristic sequence of the reference fish and the target fish, and a behavior characteristic conversion relationship is obtained.

[0194] Based on the morphological characteristic conversion relationship and the behavior characteristic conversion relationship, a characteristic parameter conversion relationship from the reference fish to the target fish is constructed.

[0195] In the embodiment, the principal component analysis method is used to extract the scale arrangement characteristic vector and the fin strip morphological characteristic vector of the reference fish and the target fish from the static phenotype parameters, wherein the scale arrangement characteristic vector represents the regularity of the scale distribution, and the fin strip morphological characteristic vector reflects the geometric characteristics of the fin strip structure.

[0196] The behavior characteristic sequence is a standardized dynamic characteristic processed by the dynamic time warping algorithm. The standardized movement trajectory eliminates the influence of the size difference of individuals, and the normalized swim bladder vibration spectrum unifies the sound intensity reference of different individuals.

[0197] The morphological characteristic conversion relationship is a linear mapping relationship between the principal components of the morphological characteristics of the reference fish and the target fish established by the canonical correlation analysis, and each element in the conversion matrix represents a conversion coefficient of the corresponding characteristic component. The behavior characteristic conversion relationship is a time sequence alignment rule between the behavior characteristic sequences of the reference fish and the target fish obtained by the dynamic time warping algorithm, and the alignment path defines the time correspondence of the behavior characteristics of different fish species. The characteristic parameter conversion relationship is a comprehensive conversion model integrating the morphological characteristic conversion relationship and the behavior characteristic conversion relationship, and realizes the standardized conversion of cross-species characteristics.

[0198] The embodiment establishes the conversion relationship of the morphological characteristics of the reference fish and the target fish through principal component analysis and canonical correlation analysis, realizes the cross-species alignment of the behavior characteristics by using the dynamic time warping algorithm, and effectively solves the characteristic difference problem between different fish species by constructing the characteristic parameter conversion relationship. The cross-species application of the fish characteristic database provides a reliable technical means, and significantly improves the universality and comparability of the characteristic parameters.

[0199] In some embodiments, the static phenotype parameters, dynamic behavior characteristics, behavior characteristic code library, mapping relationship network and characteristic parameter conversion relationship are stored in a structured manner, and a multi-dimensional characteristic database containing fish phenotype characteristics, behavior characteristics, disease characteristics and environmental parameters is constructed, including:

[0200] A storage architecture is established based on NoSQL, which includes a three-level index structure of fish individual identification, collection time stamp and environmental parameters;

[0201] The scale arrangement characteristics, fin strip morphological characteristics and supplementary phenotype characteristics in the static phenotype parameters are stored in association with the corresponding three-dimensional grid models;

[0202] The motion trajectory feature, the swim bladder vibration feature and the behavior pattern classification in the dynamic behavior feature are stored in time sequence;

[0203] The typical behavior pattern feature in the behavior feature coding library is converted into vector data, and a behavior feature association network is constructed;

[0204] The disease type, the body surface lesion feature and the behavior abnormal pattern in the mapping relationship network are associated and stored;

[0205] The feature parameter conversion relationship is stored in the form of a matrix, and a mapping relationship with a fish species classification label is established;

[0206] The various feature data in the integrated storage architecture is generated into a multi-dimensional feature database.

[0207] In the embodiment, the first level index of the three-level index structure is the fish individual identification, the second level index is the collection time stamp, and the third level index is the environmental parameter, and efficient data retrieval is achieved through the hierarchical relationship.

[0208] The scale arrangement feature, fin strip morphology feature and supplementary phenotype feature in the static phenotype parameter are associated and stored with the corresponding three-dimensional grid model, and the visualization tracing of the morphology feature is ensured.

[0209] The dynamic behavior feature is coded and organized according to the time sequence, and the time sequence evolution law of the behavior feature is retained; preferably, the behavior feature coding vector is constructed into a network structure with nodes and behavior pattern similarity as edges through the graph database technology, that is, a behavior feature association network; further, the disease diagnosis elements such as disease type, body surface lesion feature and behavior abnormal pattern in the mapping relationship network are organized through the table association mechanism of the relational database.

[0210] Preferably, the feature parameter conversion relationship stored in the form of a matrix adopts sparse matrix compression technology to optimize the storage space, and the fish species classification label mapping is realized through key-value pairs for fast query. Finally, through data fusion technology, the heterogeneous feature data is uniformly stored in a distributed database system, and a multi-dimensional feature database is obtained.

[0211] The embodiment realizes efficient organization of massive fish feature data through a three-level index structure, the three-dimensional grid model associated storage ensures the integrity of the morphology feature, the time sequence coding storage retains the dynamic characteristics of the behavior feature, the behavior feature association network enhances the analysis ability of the behavior pattern, the matrix storage of the feature parameter conversion relationship optimizes the cross-species query efficiency, and finally the multi-dimensional feature database constructed provides comprehensive data support and intelligent analysis basis for fish research.

[0212] In a second aspect, the present application also provides a live fish underwater feature database system 1, which is suitable for the method of the first aspect, and the system comprises:

[0213] The information collection module 11 is configured to collect original fish information according to a preset frequency, the original fish information including a first fish image, first acoustic information and first environmental parameters, and to pre-process the original fish information to obtain processed fish information, the pre-processing including data cleaning and spatio-temporal alignment, the processed fish information including a second fish image, second acoustic information and second environmental parameters after spatio-temporal alignment;

[0214] The feature extraction module 12 is configured to perform three-dimensional point cloud reconstruction on the second fish image to extract static phenotypic parameters, the static phenotypic parameters including fish body surface scale arrangement features and fin strip morphological features, and to perform multispectral analysis on the second fish image to obtain body surface texture reflection characteristics; to perform time domain correlation on the second fish image and the second acoustic signal to extract dynamic behavior features, the dynamic behavior features including fish body movement trajectory and swim bladder vibration frequency, and to establish a behavior feature code library containing typical behavior modes, the typical behavior modes including at least one of feeding, escaping and fighting.

[0215] The database construction module 13 is configured to construct a mapping relationship network of different disease types and body surface lesion area spatial distribution and behavior abnormality mode according to the static phenotypic parameters, the body surface texture reflection characteristics, the dynamic behavior features and the behavior feature code library; to establish a feature parameter conversion relationship between a reference fish species and a target fish species based on the static phenotypic parameters and the dynamic behavior features; and to store the static phenotypic parameters, the dynamic behavior features, the behavior feature code library, the mapping relationship network and the feature parameter conversion relationship in a structured manner to construct a multi-dimensional feature database containing fish phenotypic features, behavior features, disease features and environmental parameters.

[0216] In the embodiment, the information collection module 11 is a hardware system configured to obtain original fish information through underwater camera equipment and acoustic sensors according to a preset frequency; the feature extraction module 12 is a software system configured to extract various features from the second fish image and the second acoustic signal through computer vision and signal processing algorithms; and the database construction module 13 is a system component configured to realize multi-dimensional feature data storage and management by using a hybrid architecture of NoSQL and relational databases.

[0217] The underwater feature database system 1 of the embodiment performs the construction method of the underwater feature database of the first aspect, and specific contents refer to the foregoing technical solutions, which will not be described herein.

[0218] The underwater feature database system 1 of the embodiment realizes full-process automatic processing from data collection to feature extraction and then to database construction, and provides a reliable technical solution for fish research.

[0219] In a third aspect, the present application also provides a computer readable storage medium, which stores computer program instructions, wherein the computer program instructions, when executed by a processor, implement the method of the first aspect.

[0220] The computer program involved in the embodiment can be stored in a computer device readable storage medium, which includes but is not limited to a magnetic disk, a magnetic tape, a magnetic card, a floppy disk, a flash memory, an optical disk, an optical card, a read-only memory (ROM), a random access memory (RAM), an erasable programmable ROM (EPROM) and an electrically erasable programmable ROM (EEPROM), and the like, and also includes other biological, physical or chemical structures that can realize the same or equivalent functions as the above-mentioned storage medium, such as DNA, RNA, protein and the like units with information storage capability. In specific embodiments, the storage medium can be one of the above-mentioned medium types, or a combination of the above-mentioned medium types. In different embodiments, the computer program involved in the embodiment can be centrally stored in a single medium, or can be distributedly stored in multiple media. The storage medium containing the computer device readable storage medium can be a non-volatile memory or a random access memory. These computer device readable storage media can be built-in in the device, or can be connected with the device as an external device or part of the external device. In some embodiments, the storage medium containing the computer device readable storage medium is deployed locally; in other embodiments, the storage medium can also be deployed remotely from the processor, for example, a network-attached storage accessed via an RF circuit or an external port and a communication network, wherein the communication network can be the Internet, one or more intranets, a local area network (LAN), a wide area network (WAN), a storage area network (SAN) and the like, or a suitable combination thereof, as long as the computer device can access the storage medium. In addition, the computer program involved in the embodiment can be stored in plaintext / ciphertext form, or can be designed as training data and integrated and reorganized in the parameter state of a deep neural network or other machine learning model by means of model training.

[0221] By adopting the above technical solutions, the present application is different from the prior art and has the following beneficial effects:

[0222] The application ensures the spatio-temporal consistency of multi-source data by synchronously collecting and preprocessing the first fish image, the first acoustic information and the first environmental parameter to obtain the second fish image, the second acoustic information and the second environmental parameter after spatio-temporal alignment; the static phenotype parameters containing the scale arrangement features and fin strip morphological features of the fish body surface are extracted from the second fish image by using the three-dimensional point cloud reconstruction technology, and the body surface texture reflection characteristics obtained by the multi-spectral analysis are combined to realize comprehensive digital representation of the fish phenotype features; the dynamic behavior features containing the fish body motion trajectory and the swim bladder vibration frequency are extracted by performing time domain correlation analysis on the second fish image and the second acoustic signal, and a behavior feature coding library containing typical behavior modes is established; based on the static phenotype parameters, the body surface texture reflection characteristics, the dynamic behavior features and the behavior feature coding library, a mapping relationship network of different disease types and body surface lesion region spatial distribution and behavior abnormal mode is constructed; the cross-species feature application is realized by establishing the feature parameter conversion relationship between the benchmark fish and the target fish; finally, the static phenotype parameters, the dynamic behavior features, the behavior feature coding library, the mapping relationship network and the feature parameter conversion relationship are stored in a structured manner to construct a multi-dimensional feature database containing fish phenotype features, behavior features, disease features and environmental parameters. The above technical scheme solves the spatio-temporal consistency problem of underwater fish feature collection through collaborative analysis of optical, acoustic and environmental parameters, and significantly improves the accuracy of disease diagnosis and behavior analysis.

[0223] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of the present application, they do not limit the patent protection scope of the present application. Any technical solution obtained by replacing or modifying the equivalent structure or equivalent process based on the essential concept of the present application, using the content described in the specification and drawings of the present application, and directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, etc., are all included in the patent protection scope of the present application.

Claims

1. A method of constructing a database of underwater features of live fish, characterized by, The method comprises the following steps: Collecting original fish information according to a preset frequency, wherein the original fish information comprises a first fish image, first acoustic information and first environmental parameters; Preprocessing the original fish information to obtain processed fish information, wherein the preprocessing comprises data cleaning and space-time alignment, and the processed fish information comprises a second fish image, second acoustic information and second environmental parameters after space-time alignment; Reconstructing a three-dimensional point cloud of the second fish image to extract static phenotype parameters, wherein the static phenotype parameters comprise fish body surface scale arrangement features and fin strip morphological features, and performing multispectral analysis on the second fish image to obtain body surface texture reflection characteristics; Performing time domain correlation on the second fish image and the second acoustic signal to extract dynamic behavior features, wherein the dynamic behavior features comprise fish body movement trajectory and swim bladder vibration frequency, and establishing a behavior feature code library containing typical behavior modes, wherein the typical behavior modes comprise at least one of feeding, escaping and fighting; According to the static phenotype parameters, body surface texture reflection characteristics, dynamic behavior features and behavior feature code library, a mapping relationship network of different disease types and body surface lesion area spatial distribution and behavior abnormality modes is constructed; Based on the static phenotype parameters and dynamic behavior features, a feature parameter conversion relationship between a reference fish species and a target fish species is established; The static phenotype parameters, dynamic behavior features, behavior feature code library, mapping relationship network and feature parameter conversion relationship are stored in a structured manner to construct a multi-dimensional feature database containing fish phenotype features, behavior features, disease features and environmental parameters.

2. The method of claim 1, wherein, The method comprises the following steps: Filtering images meeting a preset clarity threshold from the first fish image as second fish images, wherein the preset clarity threshold comprises edge sharpness and contrast indicators; Extracting acoustic data in the first acoustic information matching a time point of collecting the second fish images as second acoustic information, wherein a time deviation of the matching time point does not exceed a preset synchronization threshold; Selecting environmental monitoring data corresponding to the time point of collecting the second fish images from the first environmental parameters as second environmental parameters; Establishing a time sequence correlation relationship among the second fish images, second acoustic information and second environmental parameters, so that the three have unified time coding identifiers; Performing optical compensation processing on the second fish images to eliminate image quality degradation caused by water scattering; Performing noise suppression processing on the second acoustic information to retain fish biological feature signals in an effective frequency band; Generating processed fish information, and storing a plurality of the processed fish information in a preprocessing database.

3. The method of claim 1, wherein, The method comprises the following steps: Reconstructing a three-dimensional point cloud of the second fish image to extract static phenotype parameters, wherein the static phenotype parameters comprise fish body surface scale arrangement features and fin strip morphological features selecting multi-view fish body surface images from the second fish images as three-dimensional reconstruction input data, generating dense point cloud data of the fish body surface, the dense point cloud data containing three-dimensional space coordinates and texture information; extracting fish body contour feature points from the dense point cloud data, and constructing a three-dimensional mesh model of the fish body; calibrating scale distribution areas on the three-dimensional mesh model of the fish body, and extracting scale arrangement features, the scale arrangement features including scale spacing and scale arrangement density; identifying fin strip structures in the three-dimensional mesh model, and extracting fin strip morphological features, the fin strip morphological features including size information of each fin strip, the size information of the fin strip including length, angle and relative position; storing the scale arrangement features and the fin strip morphological features in association with fish individual identification information to form structured phenotype features; and performing surface curvature analysis on the three-dimensional mesh model to extract morphological parameters of each part of the fish body as supplementary phenotype features; arranging the supplementary phenotype features and the structured phenotype features into static phenotype parameters.

4. The method of claim 1, wherein, performing multispectral analysis on the second fish images to obtain body surface texture reflection characteristics, including: separating fish body surface image data of different spectral bands from the second fish images to obtain a multispectral image data set, the spectral bands including a visible light band and a near-infrared band; performing normalization processing on the multispectral image data set to obtain a standardized multispectral image, eliminating the influence of light intensity difference on reflection characteristics; extracting reflectivity data of the back, lateral line and abdomen of the fish from the standardized multispectral image to generate reflectivity curves of each part; constructing reflectivity feature maps of different parts of the fish body based on the reflectivity curves, the reflectivity feature maps containing the correspondence between wavelength and reflection intensity; performing texture feature extraction on the standardized multispectral image, calculating gray level co-occurrence matrix features and local binary pattern features under each spectral band to obtain multispectral texture feature parameters; storing the reflectivity feature maps and the multispectral texture feature parameters in association with fish individual identification information to form structured texture features; and performing feature fusion on the multispectral texture feature parameters to extract comprehensive texture features across bands as supplementary texture features; arranging the supplementary texture features and the structured texture features into body surface texture reflection characteristics.

5. The method of claim 1, wherein, performing time domain correlation on the second fish images and the second acoustic signals to extract dynamic behavior features, and establishing a behavior feature encoding library containing typical behavior patterns, including: performing inter-frame motion analysis on the second fish image sequence to obtain fish body motion trajectory data, the fish body motion trajectory data containing position, velocity and acceleration parameters; performing time-frequency analysis on the second acoustic information to extract swim bladder vibration feature parameters, the swim bladder vibration feature parameters including vibration frequency, vibration amplitude and vibration duration; time-synchronously matching the fish body motion trajectory data and the swim bladder vibration feature parameters to establish a motion-acoustic joint feature matrix; identifying typical behavior pattern features from the motion-acoustic joint feature matrix, the typical behavior patterns including feeding behavior features, escape behavior features and fighting behavior features; Parameterize the typical behavior pattern features to generate behavior feature encoding vectors, which contain combined parameters of motion trajectory features and acoustic vibration features; Store the behavior feature encoding vectors in association with fish individual identification information to form a structured behavior feature library; Perform cluster analysis on the behavior feature encoding vectors to establish a behavior pattern classification model as a supplementary behavior feature; Organize the supplementary behavior features and the structured behavior feature library into a behavior feature encoding library.

6. The method of claim 1, wherein, According to the static phenotype parameters, body surface texture reflection characteristics, dynamic behavior features, and the behavior feature encoding library, construct a mapping relationship network of different disease types and body surface lesion area spatial distribution, behavior abnormality patterns, which includes: Extract lesion-related morphological features from the static phenotype parameters, including abnormal scale shedding area and fin damage degree; Analyze the abnormal reflection area in the body surface texture reflection characteristics to identify the spatial distribution characteristics of body surface lesions, including lesion area proportion and lesion area geometric characteristics; Screen abnormal behavior parameters from the dynamic behavior features, including feeding frequency reduction amplitude and swimming trajectory disorder degree; Perform correlation analysis on the lesion-related morphological features, body surface lesion spatial distribution characteristics, and the behavior abnormality parameters to establish a disease-phenotype-behavior three-dimensional feature space; Construct a disease type discrimination model in the three-dimensional feature space, which outputs the mapping relationship of disease type and feature combination; Correlate the disease type discrimination model with the behavior feature encoding library to generate an abnormal behavior feature set containing typical disease behavior patterns; Perform visual processing on the abnormal behavior feature set to construct a disease feature map as a supplementary diagnostic basis; Integrate the disease feature map, disease type discrimination model, and abnormal behavior feature set to form a disease feature mapping relationship network.

7. The method of claim 1, wherein, Based on the static phenotype parameters and dynamic behavior features, establish the feature parameter conversion relationship between the reference fish and the target fish, which includes: Use principal component analysis to extract morphological feature principal components of the reference fish and the target fish from the static phenotype parameters, including scale arrangement feature vectors and fin morphology feature vectors; Use dynamic time warping algorithm to extract behavior feature sequences of the reference fish and the target fish from the dynamic behavior features, including standardized motion trajectories and normalized swim bladder vibration spectra; Use canonical correlation analysis method to calculate the conversion matrix between the morphological feature principal components of the reference fish and the target fish to obtain the morphological feature conversion relationship; Apply dynamic time warping algorithm to calculate the alignment path between the behavior feature sequences of the reference fish and the target fish to obtain the behavior feature conversion relationship; Based on the morphological feature conversion relationship and the behavior feature conversion relationship, construct the feature parameter conversion relationship from the reference fish to the target fish.

8. The method of claim 1, wherein: The static phenotype parameters, dynamic behavior characteristics, behavior characteristic code library, mapping relationship network, and feature parameter conversion relationship are stored in a structured manner, and a multi-dimensional feature database containing fish phenotype characteristics, behavior characteristics, disease characteristics, and environmental parameters is constructed, including: A storage architecture based on NoSQL is established, which includes a three-level index structure of fish individual identification, collection timestamp, and environmental parameters; The scale arrangement characteristics, fin strip morphology characteristics, and supplementary phenotype characteristics in the static phenotype parameters are stored in association with corresponding three-dimensional grid models; The motion trajectory characteristics, swim bladder vibration characteristics, and behavior pattern classification in the dynamic behavior characteristics are stored in chronological order; The typical behavior pattern characteristics in the behavior characteristic code library are converted into vector data, and a behavior characteristic association network is constructed; The disease types, body surface lesion characteristics, and behavior abnormal patterns in the mapping relationship network are stored in association; The feature parameter conversion relationship is stored in matrix form, and a mapping relationship with fish species classification labels is established; The various feature data in the storage architecture are integrated to generate the multi-dimensional feature database.

9. A live fish underwater feature database system, characterized by, The system is suitable for the method of any one of claims 1-8, and the system comprises: An information collection module, configured to collect original fish information at a preset frequency, wherein the original fish information includes first fish images, first acoustic information, and first environmental parameters, and the original fish information is preprocessed to obtain processed fish information, wherein the preprocessing includes data cleaning and spatiotemporal alignment, and the processed fish information includes second fish images after spatiotemporal alignment, second acoustic information, and second environmental parameters; A feature extraction module, configured to perform three-dimensional point cloud reconstruction on the second fish images to extract static phenotype parameters, wherein the static phenotype parameters include scale arrangement characteristics and fin strip morphology characteristics of a fish body surface, and perform multispectral analysis on the second fish images to obtain body surface texture reflection characteristics; perform time domain correlation on the second fish images and second acoustic signals to extract dynamic behavior characteristics, wherein the dynamic behavior characteristics include fish body motion trajectory and swim bladder vibration frequency, and a behavior characteristic code library containing typical behavior patterns is established, wherein the typical behavior patterns include at least one of feeding, escaping, and fighting; A database construction module, configured to construct a mapping relationship network of different disease types and body surface lesion region spatial distribution and behavior abnormal patterns based on the static phenotype parameters, body surface texture reflection characteristics, dynamic behavior characteristics, and behavior characteristic code library; establish a feature parameter conversion relationship between a reference fish species and a target fish species based on the static phenotype parameters and dynamic behavior characteristics; and store the static phenotype parameters, dynamic behavior characteristics, behavior characteristic code library, mapping relationship network, and feature parameter conversion relationship in a structured manner, and construct a multi-dimensional feature database containing fish phenotype characteristics, behavior characteristics, disease characteristics, and environmental parameters.

10. A computer readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by a processor, implement the method of any one of claims 1 to 8.

Images