Insect high-efficiency identification system based on fusion of microscopic images and artificial intelligence

Abstract

The application discloses an efficient insect identification system based on microscopic image and artificial intelligence fusion, relates to the technical field of insect identification, and aims to solve the technical problem that the existing insect identification system is prone to imaging quality and identification accuracy reduction caused by environmental factors, and comprises an environment sensing and modeling module, which collects sample environment parameters in real time and constructs an environment-feature variation correlation model; an image acquisition module, which integrates an electrically adjustable focus high-resolution microscope and a multispectral imaging sensor, supports automatic adjustment of magnification, spectral acquisition channels and focusing parameters; a sample characteristic analysis module, which extracts the morphology, texture and spectral reflection characteristics of insect samples, and identifies rare species and morphological variation samples; and an intelligent decision support module, which generates image acquisition parameter combinations and preprocessing algorithm strategies based on sample characteristics and historical data. The application has the advantages of improving imaging quality and identification accuracy in the process of insect identification.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of insect identification, in particular to an efficient insect identification system based on microscopic image and artificial intelligence fusion. BACKGROUND

[0002] At present, in the field of insect microscopic image identification, the low identification accuracy caused by environmental factor interference and insufficient reliability of synthetic data has become the core bottleneck restricting the development of the technology. The morphological characteristics of insect samples are easily affected by environmental parameters (such as air pressure, temperature and humidity) to cause adaptive variation, and the traditional identification system lacks the ability of environmental perception and dynamic adjustment, which often leads to the fact that the key structure of high-altitude micro-insects is imaged fuzzy due to the contraction of the body size, and the spectral reflection characteristics are distorted due to the change of the epidermal thickness, so that the misjudgment rate of rare species is as high as 40%. In addition, in order to alleviate the problem of data scarcity of long-tail samples, the generative adversarial network (cGAN) technology is used, and the synthetic samples have problems such as biomechanical morphological distortion and spectral reflection characteristic deviation, so that the generalization ability of the model trained based on the synthetic data is significantly reduced in actual application. In view of this, the application provides an efficient insect identification system based on microscopic image and artificial intelligence fusion. SUMMARY

[0003] The application aims to provide an efficient insect identification system based on microscopic image and artificial intelligence fusion, so as to solve the technical problem that the existing insect identification system is prone to cause reduction of imaging quality and identification accuracy due to environmental factors.

[0004] To solve the above technical problems, the application provides the following technical scheme: an efficient insect identification system based on microscopic image and artificial intelligence fusion, comprising:

[0005] An environment perception and modeling module, which collects sample environmental parameters in real time and constructs an environment-feature variation correlation model;

[0006] An image acquisition module, which integrates an electrically adjustable focus high-resolution microscope and a multi-spectral imaging sensor, and supports automatic adjustment of magnification, spectral acquisition channel and focusing parameters;

[0007] A sample characteristic analysis module, which extracts the morphology, texture and spectral reflection characteristics of insect samples, and identifies rare species and morphological variation samples;

[0008] An intelligent decision support module, which generates image acquisition parameter combinations and preprocessing algorithm strategies based on sample characteristics and historical data;

[0009] A database module stores insect sample images, taxonomic features and ecological habit data, and has a data enhancement function. The data enhancement function is realized based on a generative adversarial network. Training data of the generative adversarial network is derived from historical insect sample images stored in the database module and feature data extracted by the sample characteristic analysis module.

[0010] A feature extraction and recognition module uses a deep learning model to perform hierarchical feature extraction and species recognition on images.

[0011] A synthetic data verification module verifies the biological authenticity of samples synthesized by the generative adversarial network in the database module.

[0012] Preferably, the environment perception and modeling module comprises:

[0013] An environment parameter acquisition unit is equipped with high-precision air pressure sensors, temperature sensors and humidity sensors to accurately and in real time acquire key environment parameters such as air pressure, temperature and humidity of the environment where the sample is located.

[0014] An environment-feature variation analysis submodule: This submodule deeply mines a large amount of historical data, combines with a literature knowledge base, and deeply studies the influence mechanism of environmental factors on insect morphological features. Through complex data analysis and modeling methods, an accurate quantitative relationship model between environment parameters and insect morphological feature variation is established.

[0015] A pressure-body size contraction coefficient model is constructed: Let the air pressure be , the body size contraction coefficient be , and the function be obtained by least squares fitting of historical data ;

[0016] A temperature-epidermis thickness change curve model is established: Let the temperature be , the epidermis thickness be , and the function be constructed by polynomial regression ;

[0017] A dynamic adjustment compensation unit: Real-time environment parameters acquired by the environment parameter acquisition unit and compensation parameters output by the environment-feature variation analysis submodule are timely and accurately transmitted to the intelligent decision support module. Through this data interaction, the intelligent decision support module dynamically corrects the image acquisition parameter adjustment strategy according to environmental factors, ensuring that the collected images can truly and completely reflect the morphological features of insects.

[0018] Preferably, the image acquisition module performs the following operations according to the parameter adjustment strategy generated by the intelligent decision support module:

[0019] According to the air pressure-body size contraction coefficient The intelligent decision support module can automatically and reasonably increase the magnification of the microscope;

[0020] Combining the temperature-skin thickness change curve The intelligent decision support module optimizes the channel selection of multi-spectral imaging. By calculating the influence of the skin thickness change on the spectral reflection intensity of different spectral channels, the optimal channel is determined.

[0021] Based on the real-time collected environmental parameters, the intelligent decision support module adjusts the focusing strategy of the microscope, and adopts layered scanning and image fusion technology.

[0022] Preferably, the sample characteristic analysis module comprises:

[0023] The original image obtained by the image acquisition module is preprocessed, including adaptive noise reduction, contrast enhancement and background subtraction, to improve the image quality.

[0024] A lightweight feature extraction model is used to simultaneously extract the morphological features, texture features and spectral reflection features of the insect sample, generating a sample characteristic parameter vector containing multi-dimensional information.

[0025] A built-in long-tail sample detection mechanism is used to identify rare species, extreme specialized samples or morphological variation samples by analyzing the degree of outlying of feature distribution.

[0026] Preferably, the intelligent decision support module is based on the sample characteristic parameter vector output by the sample characteristic analysis module and the environmental parameters of the environmental perception and modeling module, and performs the following operations:

[0027] A multi-objective optimization model is constructed, combining historical data and current sample characteristics to generate image acquisition parameter combinations and preprocessing algorithm strategies that adapt to the current sample.

[0028] Meta-learning technology is used to quickly adapt the collection strategy of rare samples, combining historical similar sample experience to form a dedicated adjustment scheme for long-tail distributed samples.

[0029] A dynamic strategy cache library is maintained to store and update the optimal adjustment strategy in real time, supporting efficient processing of the same type of samples

[0030] Preferably, the database module stores a large number of insect sample images, taxonomic features and ecological habit data to form a multi-dimensional knowledge base.

[0031] It has data enhancement function, using generative adversarial network, taking historical sample images and feature data extracted by sample characteristic analysis module as training set, synthesizing rare species samples, balancing training data distribution.

[0032] Construct a taxonomic knowledge graph, correlate species evolution relationships, morphological characteristics, and ecological environments, and provide semantic support for intelligent decision-making.

[0033] When storing the synthetic sample, the environmental parameter conditions simulated by the synthetic sample, the biological authenticity score and the verification results of each dimension generated by the synthetic data verification module, and the adversarial network model parameter configuration information used to generate the sample are recorded synchronously.

[0034] Preferably, the synthetic data verification module comprises:

[0035] Biomechanical verification unit: based on the biomechanical model of insect cuticle tension and material mechanical properties, the synthetic sample generated by the generative adversarial network in the database module is comprehensively verified for morphological rationality;

[0036] Spectral consistency analysis unit: using professional spectral reflectance measurement equipment, the spectral data of the real sample are obtained as a reference standard, and the spectral data of the synthetic sample in the database module are compared with the spectral data of the real sample in each wave band. Let the spectral reflectance of the real sample be , and the spectral reflectance of the synthetic sample be , calculate the spectral deviation :

[0037] ;

[0038] wherein, is the number of spectral bands, and the consistency degree of the spectral characteristics of the synthetic sample and the real sample is evaluated;

[0039] Multi-modal fusion verification submodule: the verification results of the biomechanical verification unit and the analysis results of the spectral consistency analysis unit are fused, and a weighted summation algorithm is used to generate the biological authenticity score of the synthetic sample :

[0040] ;

[0041] wherein, and are weight coefficients, and are pre-set maximum allowed difference values, and when the score is lower than the pre-set threshold , the parameter optimization process of the generative adversarial network in the database module is triggered, and the sample is required to be regenerated.

[0042] Preferably, when the biomechanical verification unit adopts the finite element analysis method, the following differentiated verification operations are further performed:

[0043] Precise assignment of material parameters: Based on an insect epidermal material property database, precise elastic modulus is assigned to the wing vein structure of the synthetic sample. Poisson's ratio The mechanical parameters are used to ensure the accuracy of the basic data for finite element analysis.

[0044] Structural deformation depth simulation: Using finite element analysis technology, the fin structure is divided into multiple elements, and the element stiffness matrix is ​​solved. and overall stiffness matrix The morphological changes of the wing vein structure under different stress scenarios were simulated under the action of epidermal tension to obtain a more comprehensive bending angle and radius of curvature.

[0045] Strict threshold determination: The calculated wing vein structure feature parameters are rigorously compared with the corresponding morphological parameters of the real sample, and multiple levels of difference thresholds are set, with a primary threshold... For initial screening, advanced threshold Used for final determination, when the bending angle difference Exceeding the advanced threshold ,Right now The results indicate that the wing vein structure of the synthesized sample has serious biomechanical inconsistencies, requiring a significant adjustment to the synthesis process or the regeneration of the sample.

[0046] Preferably, the feature extraction and recognition module employs the following mechanism when processing samples:

[0047] Biological authenticity scoring based on synthetic samples Dynamically adjust the attention weights of the feature extraction network;

[0048] For synthetic samples with low biological authenticity scores, reduce the priority of identifying suspicious features.

[0049] By combining the environmental parameters of the synthetic sample with the simulated conditions, the environment-feature variation model is invoked to compensate and correct the feature extraction results.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1. This invention systematically constructs a complete link of "parameter acquisition - variation modeling - strategy compensation" through environmental perception and dynamic adjustment mechanisms. The environmental perception and modeling module monitors environmental parameters in real time, conducts in-depth research on the influence mechanism of environmental factors on insect morphological characteristics, establishes an environmental-feature variation quantification model, accurately calculates compensation parameters, and the image acquisition module dynamically adjusts the acquisition strategy accordingly. This effectively solves the problems of imaging quality and identification accuracy caused by environmental factors, and can clearly and completely present the key identification features of insect samples.

[0052] 2. This invention also relies on a multi-dimensional verification system for synthetic data, which systematically breaks through the technical bottleneck of the reliability of synthetic samples. The synthetic data verification module rigorously verifies the cGAN synthetic samples from both biomechanical and spectral reflectance dimensions. The biomechanical verification unit simulates the stress deformation of the sample structure and compares the morphological parameters of the real sample with those of the synthetic sample to reduce the risk of misjudging the structural morphology. The spectral consistency analysis unit compares the spectral reflectance band by band to reduce the spectral deviation between the synthetic sample and the real sample. This mechanism ensures that the synthetic samples in the database have high biological authenticity and provides a high-quality data foundation for model training.

[0053] 3. This invention utilizes a knowledge graph-driven intelligent evolutionary architecture to continuously optimize the system's identification capabilities. The knowledge update feedback loop transmits information such as the environmental-feature variation relationship and the results of synthetic sample verification to the knowledge graph construction unit of the database module in real time. Advanced algorithms are used to dynamically update knowledge nodes. When new environmental adaptive variations occur or synthetic data verification standards are updated, the system can automatically adjust relevant model parameters and generation strategies. This enables the system to continuously optimize identification strategies when facing complex environmental samples and new long-tailed species, forming a closed-loop ecosystem of "data collection - intelligent decision-making - knowledge evolution". Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0055] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0056] Example 1, such as Figure 1 As shown, this invention provides a highly efficient insect identification system based on the fusion of microscopic images and artificial intelligence, comprising:

[0057] The environmental perception and modeling module collects sample environmental parameters in real time and constructs an environment-feature variation correlation model.

[0058] The image acquisition module integrates a motorized adjustable-focus high-resolution microscope and a multispectral imaging sensor, supporting automatic adjustment of magnification, spectral acquisition channels, and focusing parameters.

[0059] The sample characteristic analysis module extracts the morphological, texture, and spectral reflectance characteristics of insect samples to identify rare species and morphologically varied samples.

[0060] The intelligent decision support module generates image acquisition parameter combinations and preprocessing algorithm strategies based on sample characteristics and historical data.

[0061] A database module stores insect sample images, taxonomic features and ecological habit data, and has a data enhancement function, which is realized based on a generative adversarial network, and the training data of the generative adversarial network is derived from historical insect sample images stored in the database module and feature data extracted by the sample characteristic analysis module;

[0062] A feature extraction and recognition module uses a deep learning model to perform hierarchical feature extraction and species recognition on images;

[0063] A synthetic data verification module verifies the biological authenticity of samples synthesized by the generative adversarial network in the database module.

[0064] In an embodiment of the present application, the environment perception and modeling module comprises:

[0065] An environment parameter acquisition unit is equipped with high-precision air pressure sensors, temperature sensors and humidity sensors to collect key environmental parameters such as air pressure, temperature and humidity of the environment where the sample is located in real time and accurately. These sensors are professionally calibrated to ensure the accuracy and reliability of the collected data, providing a solid data foundation for subsequent analysis;

[0066] An environment-feature variation analysis submodule: this submodule deeply mines a large amount of historical data, combines with authoritative literature knowledge base, deeply studies the influence mechanism of environmental factors on insect morphological features, and through complex data analysis and modeling methods, establishes an accurate quantitative relationship model between environmental parameters and insect morphological feature variation;

[0067] Constructing an air pressure-body size contraction coefficient model: let the air pressure be , the body size contraction coefficient be , the function be obtained by least squares fitting of historical data , and the specific formula be:

[0068] ;

[0069] Wherein, is the actual observed body size contraction value, is the corresponding air pressure value, is the sample quantity;

[0070] Establishing a temperature-epidermis thickness change curve model: let the temperature be , the epidermis thickness be , and the function be constructed by polynomial regression : ;

[0071] Wherein, is the polynomial coefficient, is the polynomial order, and the optimal value is determined by maximum likelihood estimation.

[0072] The dynamic adjustment compensation unit timely and accurately synchronously transmits the real-time environmental parameters acquired by the environmental parameter acquisition unit and the compensation parameters output by the environmental-feature variation analysis submodule to the intelligent decision support module, and through the data interaction, the intelligent decision support module dynamically corrects the image acquisition parameter adjustment strategy according to the environmental factors, and ensures that the collected images can truly and completely reflect the morphological characteristics of insects.

[0073] In the embodiment of the application, the image acquisition module performs the following operations according to the parameter adjustment strategy generated by the intelligent decision support module:

[0074] According to the air pressure-body size contraction coefficient provided by the environmental-feature variation analysis submodule The intelligent decision support module can automatically and reasonably increase the magnification of the microscope;

[0075] Let the original magnification be The adjusted magnification is The calculation formula is: ;

[0076] This operation is to compensate for the reduction in structural size caused by the contraction of the insect body size due to the decrease in air pressure in a high-altitude environment, so as to ensure that the microstructure of the insect can be clearly presented in the microscopic image, and avoid missing the key identification features due to insufficient imaging scale;

[0077] In combination with the temperature-epidermis thickness change curve The intelligent decision support module optimizes the channel selection of multi-spectral imaging, calculates the influence of epidermis thickness change on spectral reflectance intensity under different spectral channels, determines the optimal channel, and sets the relationship between spectral reflectance intensity and epidermis thickness as , wherein is the spectral wavelength, and the channel is selected by maximizing the objective function :

[0078] ;

[0079] Since the temperature change in a high-altitude environment will cause the change of the epidermis thickness of the insect, and then affect its spectral reflectance characteristics, by selecting the appropriate spectral channel, the contrast of the spectral reflectance characteristics weakened due to the change of the epidermis thickness can be effectively enhanced, so that the feature information of the insect is more prominent in the image, and the subsequent feature extraction and identification are facilitated;

[0080] Based on the real-time collected environmental parameters, the intelligent decision support module adjusts the focusing strategy of the microscope, adopts the layered scanning and image fusion technology, and sets the sample depth as Different depth images are The final image is obtained by a weighted fusion algorithm :

[0081] ;

[0082] wherein, is a weight coefficient determined according to sample depth and imaging quality, so as to compensate for the influence of physical deformation of the sample caused by changes in atmospheric pressure in a high-altitude environment on the depth of field, clear images of the sample at different depths are obtained through layered scanning, and then the images are fused to obtain a complete image with sufficient depth of field, so that each part of the insect sample can be clearly imaged.

[0083] In an embodiment of the present application, the sample characteristic analysis module comprises:

[0084] The original image obtained by the image acquisition module is preprocessed, including adaptive noise reduction, contrast enhancement and background subtraction, to improve the image quality;

[0085] A lightweight feature extraction model is used to synchronously extract morphological features, texture features and spectral reflection features of the insect sample, to generate a sample characteristic parameter vector containing multi-dimensional information;

[0086] A long-tail sample detection mechanism is built in, which identifies rare species, extremely specialized samples or morphological variation samples by analyzing the degree of outlying of feature distribution.

[0087] In an embodiment of the present application, the intelligent decision support module executes the following operations based on the sample characteristic parameter vector output by the sample characteristic analysis module and the environmental parameters of the environmental perception and modeling module:

[0088] A multi-objective optimization model is constructed, combining historical data and current sample characteristics to generate image acquisition parameter combinations and preprocessing algorithm strategies that adapt to the current sample;

[0089] Meta-learning technology is used to quickly adapt the collection strategy of rare samples, combining historical similar sample experience to form an exclusive adjustment scheme for long-tail distribution samples;

[0090] A dynamic strategy cache library is maintained to store and update optimal adjustment strategies in real time, supporting efficient processing of the same type of samples

[0091] In an embodiment of the present application, the database module stores massive insect sample images, taxonomic features and ecological habit data to form a multi-dimensional knowledge base;

[0092] It has a data enhancement function, using a generative adversarial network, using historical sample images and feature data extracted by the sample characteristic analysis module as a training set to synthesize rare species samples and balance the distribution of training data;

[0093] Construct a taxonomic knowledge graph, associate species evolution relationship, morphological characteristics and ecological environment, and provide semantic support for intelligent decision-making;

[0094] When storing the synthetic sample, the environmental parameter conditions simulated by the synthetic sample, the biological authenticity score and the verification results of each dimension generated by the synthetic data verification module, and the adversarial network model parameter configuration information used to generate the sample are recorded synchronously.

[0095] In the embodiment of the present application, the synthetic data verification module comprises:

[0096] The biomechanics verification unit: based on the biomechanics model of insect cuticle tension and material mechanics characteristics, the sample synthesized by the generative adversarial network in the database module is comprehensively verified for morphological rationality, by setting the displacement vector of the wing vein structure as by solving the balance equation: wherein, is the stiffness matrix, is the external force vector, and then calculating the bending angle and the radius of curvature ;

[0097] The bending angle of the real sample is , the bending angle of the synthetic sample is , and the difference degree is calculated :

[0098] ;

[0099] Whether the structure and morphology of the synthetic sample conform to the biomechanical characteristics of insects is judged in this way;

[0100] The spectral consistency analysis unit: using professional spectral reflectance measurement equipment, the spectral data of the real sample are obtained as a reference standard, the spectral data of the synthetic sample in the database module are compared with the real sample spectrum wave by wave, the real sample spectral reflectance is , the synthetic sample spectral reflectance is , and the spectral deviation is calculated :

[0101] ;

[0102] wherein, is the number of spectral bands, and the consistency degree of the spectral characteristics of the synthetic sample and the real sample is evaluated;

[0103] A multi-modal fusion verification sub-module: fuses the verification results of the biomechanical verification unit and the analysis results of the spectral consistency analysis unit, and generates a biological authenticity score of the synthetic sample by using a weighted summation algorithm :

[0104] ;

[0105] wherein, and are weight coefficients, and are preset maximum allowable difference values, when the score is lower than a preset threshold , a parameter optimization process of a generative adversarial network in the database module is triggered, and the sample is required to be regenerated.

[0106] In an embodiment of the present application, when the biomechanical verification unit adopts a finite element analysis method, the following differentiated verification operations are further performed:

[0107] Material parameter accurate assignment: based on the insect cuticle material attribute database, accurate elastic modulus , Poisson's ratio mechanical parameters are assigned to the wing vein structure of the synthetic sample, ensuring the accuracy of the basic data of the finite element analysis;

[0108] Structure deformation depth simulation: using finite element analysis technology, the wing vein structure is divided into multiple units, by solving the unit stiffness matrix and the overall stiffness matrix , the morphological change process of the wing vein structure under the action of cuticle tension in different stress scenarios is simulated, and more comprehensive bending angles and curvature radii are obtained;

[0109] Strict threshold judgment: strictly compare the calculated wing vein structure feature parameters with the corresponding morphological parameters of the real sample, set multiple difference thresholds, the primary threshold is used for preliminary screening, and the advanced threshold is used for final determination, when the bending angle difference exceeds the advanced threshold , that is, , it is determined that the wing vein structure of the synthetic sample has serious biomechanical irrationality, and the synthesis process needs to be deeply adjusted or the sample needs to be regenerated.

[0110] In an embodiment of the present application, the feature extraction and recognition module adopts the following mechanism when processing samples:

[0111] Based on the biological authenticity score of the synthetic sample , the attention weight of the feature extraction network is dynamically adjusted, the original attention weight is , the adjusted weight is , the adjustment is performed through the function

[0112] ;

[0113] For synthetic samples with low biological authenticity scores, the identification priority of suspicious features in them is reduced;

[0114] In combination with the environmental parameter simulation conditions of the synthetic sample, the environmental-feature variation model is called to compensate and correct the feature extraction result, the original feature vector is , the corrected feature vector is , the correction is performed through the function

[0115] ;

[0116] The influence caused by environmental simulation deviation is eliminated, and the accuracy of identification is improved.

[0117] In the embodiments of the present application, further comprising:

[0118] Knowledge update feedback link: the detailed analysis results of the environment perception and modeling module and the synthetic data verification module, such as the environmental-feature variation function, the biological authenticity score of the synthetic sample, and the verification dimension result, are timely and accurately synchronized to the knowledge graph construction unit of the database module;

[0119] The knowledge graph construction unit updates the environmental-feature variation relationship and the biological authenticity evaluation standard of the synthetic sample in the knowledge graph based on the feedback data, adopts a graph convolution network algorithm, and comprehensively updates the key knowledge nodes in the knowledge graph, and the knowledge graph is:

[0120] ;

[0121] Wherein, the node represents a knowledge node, the edge represents the relationship between nodes, and the node feature vector is updated through iteration;

[0122] ;

[0123] Wherein, is the number of network layers, is the neighbor node set of the node , is the weight matrix, is the bias vector, and represents the iteration of the node​​ In the layer network, the feature vector of the neighbor node , The activation function realizes dynamic evolution of the system knowledge hierarchy.

[0124] The embodiments of the present application are disclosed, but not limited to the preferred embodiments, and the ordinary skilled in the art can easily understand the spirit of the present application and make different inferences and changes according to the above embodiments, as long as they do not deviate from the spirit of the present application, which are within the protection scope of the present application.

Claims

1. An efficient insect identification system based on the fusion of microscopic images and artificial intelligence, characterized in that, Comprise: An environment perception and modeling module that collects sample environment parameters in real time and constructs an environment-feature variation correlation model; An image acquisition module that integrates an electrically adjustable focus high-resolution microscope and a multi-spectral imaging sensor, supports automatic adjustment of magnification, spectral acquisition channel and focusing parameters; A sample characteristic analysis module that extracts the morphology, texture and spectral reflection characteristics of insect samples, and identifies rare species and morphological variation samples; An intelligent decision support module that generates image acquisition parameter combinations and preprocessing algorithm strategies based on sample characteristics and historical data; A database module that stores insect sample images, taxonomic characteristics and ecological habit data, and has a data enhancement function, which is realized based on a generative adversarial network, and the training data of the generative adversarial network comes from historical insect sample images stored in the database module and feature data extracted by the sample characteristic analysis module; A feature extraction and recognition module that uses a deep learning model to perform hierarchical feature extraction and species recognition on images; A synthetic data verification module that verifies the biological authenticity of samples synthesized by the generative adversarial network in the database module; The synthetic data verification module comprises: A biomechanics verification unit that performs comprehensive morphological rationality verification on samples synthesized by the generative adversarial network in the database module based on a biomechanics model of insect cuticle tension and material mechanics characteristics; Spectrum consistency analysis unit: using professional spectral reflectance measurement equipment, obtaining the spectral data of the real sample as a reference standard, comparing the spectral data of the synthetic sample in the database module with the real sample spectrum wave by wave, setting the real sample spectral reflectance as , the synthetic sample spectral reflectance as , and calculating the spectral deviation : ； wherein, is the number of spectral bands, and the consistency degree of the synthesized sample spectral features with the real sample is evaluated. The multimodal fusion verification submodule fuses the verification result of the biomechanical verification unit and the analysis result of the spectral consistency analysis unit, and generates a biological authenticity score of the synthetic sample by using a weighted summation algorithm : ； wherein, and are weight coefficients, and are preset maximum allowable difference values, when the score is lower than a preset threshold , a parameter optimization process of the generative adversarial network in the database module is triggered, and the sample is required to be regenerated.

2. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 1, characterized in that, The environment perception and modeling module comprises: An environment parameter acquisition unit equipped with high-precision air pressure sensors, temperature sensors and humidity sensors that can accurately collect key environmental parameters such as air pressure, temperature and humidity of the environment where the sample is located in real time; An environment-feature variation analysis submodule that deeply mines a large amount of historical data, combines with a literature knowledge base, and deeply studies the influence mechanism of environmental factors on insect morphological characteristics, and through complex data analysis and modeling methods, establishes an accurate quantitative relationship model between environmental parameters and insect morphological characteristic variations; Constructing the model of air pressure-body shrinkage coefficient: let air pressure be , body shrinkage coefficient be , and the function of be obtained by least square method fitting historical data; Establishing temperature-epidermis thickness change curve model: let temperature be , epidermis thickness be , and build function by polynomial regression; A dynamic adjustment compensation unit that synchronously transmits real-time environmental parameters obtained by the environment parameter acquisition unit and compensation parameters output by the environment-feature variation analysis submodule to the intelligent decision support module in a timely and accurate manner, and through this data interaction, the intelligent decision support module dynamically corrects the image acquisition parameter adjustment strategy according to environmental factors to ensure that the collected images can truly and completely reflect the morphological characteristics of insects.

3. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 2, characterized in that, The image acquisition module performs the following operations according to the parameter adjustment strategy generated by the intelligent decision support module: a barometric body shrinkage coefficient provided by the environmental-feature variation analysis submodule The intelligent decision support module can automatically and reasonably increase the magnification of the microscope; Temperature-skin thickness change curve The intelligent decision support module optimizes the channel selection of multispectral imaging. By calculating the influence of skin thickness change on spectral reflectance intensity under different spectral channels, the optimal channel is determined. Based on the real-time collected environmental parameters, the intelligent decision support module adjusts the focusing strategy of the microscope using hierarchical scanning and image fusion technology.

4. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 3, characterized in that, The sample characteristic analysis module comprises: Pretreats the original images obtained by the image acquisition module, including adaptive noise reduction, contrast enhancement and background subtraction, to improve image quality; Uses a lightweight feature extraction model to synchronously extract the morphological characteristics, texture characteristics and spectral reflection characteristics of insect samples, and generates a sample characteristic parameter vector containing multi-dimensional information; Has a built-in long-tail sample detection mechanism that identifies rare species, extremely specialized samples or morphological variation samples by analyzing the outlying degree of feature distribution.

5. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 4, characterized in that, The intelligent decision support module executes the following operations based on the sample characteristic parameter vector output by the sample characteristic analysis module and the environmental parameters of the environmental perception and modeling module: A multi-objective optimization model is constructed to generate image acquisition parameter combinations and preprocessing algorithm strategies that adapt to the current sample by combining historical data and current sample characteristics; Meta-learning techniques are used to quickly adapt the collection strategies of rare samples, and historical similar sample experience is combined to form exclusive adjustment schemes for long-tail distribution samples; A dynamic strategy cache library is maintained to store and update optimal adjustment strategies in real time, supporting efficient processing of the same type of samples.

6. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 5, characterized in that, The database module stores massive insect sample images, taxonomic characteristics, and ecological habit data to form a multi-dimensional knowledge base; It has data enhancement function, using the generated adversarial network, taking the historical sample images and the feature data extracted by the sample characteristic analysis module as the training set, synthesizing rare species samples, and balancing the distribution of training data; A taxonomic knowledge graph is constructed to associate species evolution relationships, morphological characteristics, and ecological environments to provide semantic support for intelligent decision-making; When storing the synthesized samples, the environmental parameter conditions simulated by the synthesized samples, the biological authenticity scores generated by the synthesized data verification module, the verification results in each dimension, and the adversarial network model parameter configuration information used to generate the samples are recorded synchronously.

7. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 6, characterized in that, When the biomechanics verification unit uses the finite element analysis method, the following differentiated verification operations are further executed: Material parameter accurate assignment: based on the insect cuticle material attribute database, the accurate elastic modulus of the wing vein structure of the synthetic sample is assigned , Poisson's ratio mechanical parameters, ensuring the accuracy of the basic data of finite element analysis; Structure deformation depth simulation: using finite element analysis technology, the vein structure is divided into multiple units, by solving the unit stiffness matrix and the overall stiffness matrix , simulate the morphological change process of the vein structure under the action of skin tension in different stress scenarios, and obtain more comprehensive bending angle and curvature radius; Strict threshold judgment: Strictly compare the calculated wing vein structure feature parameters with the corresponding morphological parameters of the real sample, set multiple difference thresholds, the primary threshold for preliminary screening, the advanced threshold for final judgment, when the difference of bending angle exceeds the advanced threshold , that is , it is determined that the wing vein structure of the synthetic sample has serious biomechanical irrationality, and the synthesis process needs to be deeply adjusted or the sample needs to be regenerated.

8. The high-efficiency insect identification system based on the fusion of microscopic images and artificial intelligence according to claim 7, characterized in that, The feature extraction and recognition module uses the following mechanisms when processing samples: Biological plausibility scoring based on synthetic samples dynamically adjust the attention weights of the feature extraction network; For synthesized samples with low biological authenticity scores, the recognition priority of suspicious features in them is reduced; Combined with the environmental parameter simulation conditions of the synthesized samples, the environmental-feature variation model is called to compensate and correct the feature extraction results.

Images