Quantitative method and system for spatial distribution of key species population in complex ecological system

Abstract

The application provides a key species population spatial distribution quantification method and system in a complex ecological system, the method comprising: performing feature extraction on biological and environmental data collected in a target ecological region, fusing the extracted multi-modal features to generate a comprehensive feature vector; analyzing the comprehensive feature vector based on a target large model to identify species in the target ecological region and obtain species identification information of each species; quantifying the spatial distribution of the species population based on the species identification information and the comprehensive feature vector corresponding to the target ecological region to generate species population quantification information including at least a population distribution map, a population density map, and a population dynamic change map; and generating a quantification report and displaying the report based on the species population quantification information and the collected original data. The application can intelligently, efficiently and accurately identify species based on a large model, and can intuitively display the species distribution and dynamic change and provide a quantification report, thereby providing a strong basis for ecological protection and management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecology, and in particular to a method and system for quantifying spatial distribution of key species population in a complex ecological system. BACKGROUND

[0002] With the rapid development of social economy and the continuous growth of population, the influence of human beings on the ecological system is increasingly significant. The health and stability of the ecological system are closely related to the development of human society. A good ecological system not only guarantees the normal function and biodiversity of the ecological system, but also provides rich ecological system services for human beings and supports the development of economy, society and culture.

[0003] Species diversity is a key factor for the health and stability of the ecological system, and plays a crucial role in maintaining ecological balance, providing ecological system services and ensuring the sustainable development of human society. The spatial distribution information of key species population is of great significance for ecological protection, resource management and environmental planning. Through scientific investigation and monitoring, the species distribution can be understood, and more effective protection and management measures can be developed.

[0004] Currently, there are many limitations in investigating the species distribution by using traditional methods, such as low data collection efficiency, uneven data quality, limited coverage of collected data, poor data timeliness, difficulty in data integration, interference with the ecological environment during the collection process, susceptibility to environmental factors, complicated data processing, difficulty in accurate quantitative analysis of species distribution, and high cost. The above-mentioned disadvantages limit the depth and breadth of species distribution research, and affect the scientificity and effectiveness of ecological protection and management decision-making.

[0005] Moreover, in the prior art, when a large model is used for species identification, the species identification large model has obvious deficiencies in cross-modal data fusion, field adaptability and edge deployment, such as lack of hierarchical alignment mechanism in multi-modal pre-training, resulting in cross-modal semantic gap problem. SUMMARY

[0006] In view of the above problems, the embodiments of the present application provide a method and system for quantifying spatial distribution of key species population in a complex ecological system, which overcomes the above problems or at least partially solves the above problems.

[0007] In a first aspect, the embodiments of the present application provide a method for quantifying spatial distribution of key species population in a complex ecological system, comprising:

[0008] The biological data and environmental data collected in the target ecological region are subjected to feature extraction, and the extracted multi-modal features are fused to generate a comprehensive feature vector, wherein the biological data at least includes species image data, species audio data and species positioning data, and the environmental data at least includes meteorological data and natural resource data;

[0009] The comprehensive feature vector is analyzed based on a target large model to identify the species in the target ecological region and obtain species identification information of each species in the target ecological region;

[0010] Based on the species identification information corresponding to the target ecological region and the comprehensive feature vector, the spatial distribution of the species population is quantified to generate species population quantification information, wherein the species population quantification information at least includes a population distribution map, a population density map and a population dynamic change map;

[0011] Based on the species population quantification information and the collected original data, a quantitative report describing the distribution characteristics of the species population is generated and displayed.

[0012] In a second aspect, the embodiments of the present application provide a system for quantifying the spatial distribution of key species population in a complex ecological system, comprising:

[0013] The extraction generation module is configured to extract features from biological data and environmental data collected in a target ecological region, and fuse the extracted multi-modal features to generate a comprehensive feature vector, wherein the biological data at least includes species image data, species audio data and species positioning data, and the environmental data at least includes meteorological data and natural resource data;

[0014] The analysis and identification module is configured to analyze the comprehensive feature vector based on a target large model to identify the species in the target ecological region and obtain species identification information of each species in the target ecological region;

[0015] The quantification generation module is configured to quantify the spatial distribution of the species population based on the species identification information corresponding to the target ecological region and the comprehensive feature vector to generate species population quantification information, wherein the species population quantification information at least includes a population distribution map, a population density map and a population dynamic change map;

[0016] The generation and display module is configured to generate a quantitative report describing the distribution characteristics of the species population based on the species population quantification information and the collected original data, and display the report.

[0017] The technical scheme of the embodiment of the present application extracts the biological data and environmental data collected in the target ecological region, fuses the extracted multi-modal features to generate a comprehensive feature vector, analyzes the comprehensive feature vector based on the target large model, and obtains the species identification information of each species in the target ecological region. Based on the large model technology, intelligent and efficient species identification can be realized, comprehensive feature representation can be provided by using multi-modal feature fusion, and the accuracy of species identification can be improved. By analyzing the species identification information and the comprehensive feature vector to generate species population quantitative information, and combining the species population quantitative information and the original data to generate a quantitative report, the distribution and dynamic changes of the species can be intuitively displayed, and a comprehensive and accurate high-quality quantitative report can be provided, thereby providing a strong basis for ecological protection and management. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A schematic diagram of a method for quantifying the spatial distribution of key species populations in a complex ecological system is shown.

[0019] Figure 2 A feature fusion flowchart based on the extracted features is shown.

[0020] Figure 3 A whole implementation flowchart of the method for quantifying the spatial distribution of key species populations in a complex ecological system is shown.

[0021] Figure 4 A schematic diagram of a system for quantifying the spatial distribution of key species populations in a complex ecological system is shown.

[0022] Figure 5 A schematic diagram of an electronic device structure is shown. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0024] It should be understood that the reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described is included in at least one embodiment of the application. Therefore, appearances of "in one embodiment" or "in an embodiment" at various places throughout the specification are not necessarily referring to the same embodiment. Furthermore, various particular features, structures, or characteristics can be combined in one or more embodiments.

[0025] In various embodiments of the present application, it should be understood that the size of the serial number of the following processes does not mean the order of execution, and the execution order of the processes should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0026] The embodiments of the present application provide a method for quantifying the spatial distribution of key species population in a complex ecological system, as shown in the method for quantifying the spatial distribution of key species population in a complex ecological system, comprising the steps of: Figure 1 As shown in the method for quantifying the spatial distribution of key species population in a complex ecological system, comprising the steps of:

[0027] In step 101, the biological data and environmental data collected in the target ecological region are subjected to feature extraction, and the extracted multi-modal features are fused to generate a comprehensive feature vector. The biological data at least includes species image data, species audio data and species positioning data, and the environmental data at least includes meteorological data and natural resource data.

[0028] The biological data and environmental data of the target ecological region are collected by using a collection device. For example, the collection device can collect data according to a set time, and the collection device is, for example, an unmanned device (drone, unmanned ship) equipped with various sensors. By using the collection device to collect data, the coverage range and data timeliness of the collected data can be ensured while improving the data collection efficiency and reducing the labor cost.

[0029] The collected biological data at least includes species image data, species audio data and species positioning data. The collected environmental data at least includes meteorological data and natural resource data, and the natural resource data, for example, includes land resource data (land use type, soil quality), water resource data, vegetation coverage, etc. Specifically, the environmental data includes, but is not limited to, temperature, humidity, soil condition, light intensity, water resource condition, air pressure, wind speed, season, vegetation coverage, etc. and other data related to the living environment of the species.

[0030] After data collection based on the collection device, the collected biological data and environmental data are subjected to feature extraction, and the extracted multi-modal features are subjected to feature fusion to generate a comprehensive feature vector. The multi-modal features include biological features and environmental features, and the biological features include image features extracted from image data, audio features extracted from audio data, and positioning features extracted from positioning data. The comprehensive feature vector generated based on feature fusion contains rich information and can comprehensively reflect the characteristics of the species population under specific time and space conditions.

[0031] Step 102, based on the target large model analyzing the comprehensive feature vector, identifying the species in the target ecological region, and obtaining the species identification information of each species in the target ecological region.

[0032] After determining the comprehensive feature vector that can comprehensively reflect the characteristics of the species population based on feature fusion, the comprehensive feature vector is subjected to reasoning analysis based on the target large model capable of species identification, and the species identification information of each species in the target ecological region output by the target large model is obtained to utilize the large model for intelligent species identification.

[0033] By identifying the species in the target ecological region based on the large model technology, large-scale data can be automatically processed, human intervention can be reduced, and species identification can be completed in a short time, ensuring identification efficiency while achieving intelligent species identification. The fusion of multi-modal features can improve the robustness of the model and reduce the impact of noise and uncertainty of single modal data on the identification results.

[0034] Step 103, based on the species identification information corresponding to the target ecological region and the comprehensive feature vector, quantifying the spatial distribution of the species population, generating species population quantification information, and the species population quantification information at least includes population distribution map, population density map and population dynamic change map.

[0035] After identifying the species in the target ecological region based on the target large model to obtain the species identification information of each species in the target ecological region, the species identification information and the comprehensive feature vector are comprehensively analyzed to quantify the spatial distribution of the species population and generate species population quantification information.

[0036] The species identification information provided by the target large model directly reflects the species situation contained in the target ecological region, and in combination with time and space information, it can also reflect the distribution of the species. The comprehensive feature vector contains features of image, audio, positioning and environmental data, which can provide more rich background information such as species living environment and geographical location, which can help understand the relationship between species distribution and environmental factors. By combining the species identification information and the comprehensive feature vector, the species distribution situation can be more comprehensively understood, including the species, quantity, distribution range and relationship with environmental factors.

[0037] The species population quantification information generated by combining the species recognition information and the comprehensive feature vector includes at least a population distribution map, a population density map, and a population dynamic change map. The population distribution map can visually display the distribution of the species in the geographical space, helping to identify the habitat and activity range of the species; the population density map is used to display the density distribution of the species in different regions, helping to assess the survival status and ecological pressure of the species; and the population dynamic change map is used to display the distribution changes of the species at different time points, helping to predict the future trend of the species and provide scientific basis for ecological protection and management.

[0038] In generating the population distribution map, the identified species position information is visualized to display the distribution of the species in the geographical space, such as using GIS (Geographic Information System) software to plot the latitude and longitude information of the species on a map; in generating the population density map, the density distribution of the species in different regions is calculated and visualized; and in generating the population dynamic change map, the distribution changes of the species at different time points are analyzed and visualized.

[0039] Step 104, based on the species population quantification information and the collected raw data, a quantitative report describing the distribution characteristics of the species population is generated and displayed.

[0040] After quantifying the spatial distribution of the species population and generating the species population quantification information, a comprehensive analysis is performed based on the species population quantification information and the raw data provided by the collection device, and a quantitative report is generated and displayed. The generated quantitative report can relatively comprehensively and accurately describe the distribution characteristics of the species population.

[0041] The population distribution map, population density map, and population dynamic change map included in the species population quantification information can visually display the distribution and dynamic changes of the species in the geographical space, providing the distribution characteristics of the species in different regions and at different times. The raw data provided by the collection device can provide more detailed information. The raw data can be used to verify the accuracy of the quantification results, such as by checking the original image and audio data to confirm the correctness of the species recognition; the detailed background information provided by the raw data can help understand the reasons for the species distribution, such as by analyzing environmental data (such as water quality, soil quality, and weather data) to understand the relationship between the species distribution and environmental factors. By combining the species population quantification information and the raw data to generate a quantitative report, a more comprehensive, accurate, and in-depth analysis can be provided, enhancing the accuracy and reliability of the analysis to provide high-quality quantitative reports.

[0042] The method for quantifying the spatial distribution of the population of key species in a complex ecological system provided by the embodiments of the present application extracts features from biological data and environmental data collected in a target ecological region, fuses the extracted multi-modal features to generate a comprehensive feature vector, analyzes the comprehensive feature vector based on a target large model, and obtains species identification information of each species in the target ecological region. The species identification information can be used to realize intelligent and efficient species identification based on large model technology, provide comprehensive feature representation through multi-modal feature fusion, and thus improve the accuracy of species identification. The species population quantification information is generated by analyzing the species identification information and the comprehensive feature vector, and the quantification report is generated by combining the species population quantification information and the original data, which can intuitively show the distribution and dynamic changes of the species and provide comprehensive and accurate high-quality quantification reports, thereby providing a strong basis for ecological protection and management.

[0043] The scheme for generating a comprehensive feature vector based on feature fusion is introduced as follows. When extracting features from biological data and environmental data collected in a target ecological region and fusing the extracted multi-modal features to generate a comprehensive feature vector, the following steps are included.

[0044] Data preprocessing is performed on the collected biological data and environmental data.

[0045] Multi-modal features are extracted from the preprocessed data, including biological features and environmental features. The biological features include image features, audio features, and positioning features.

[0046] The multi-modal features are time and space aligned.

[0047] The aligned image features, audio features, positioning features, and environmental features are combined with corresponding weights for feature splicing to generate a comprehensive feature vector.

[0048] The weights corresponding to each modal feature are dynamically adjusted based on the feature contribution degree.

[0049] Data preprocessing is required before feature extraction of the collected data. When preprocessing biological data, for example, the following methods are used: image denoising, image enhancement, and image standardization processing are performed on species image data to obtain the image to be analyzed; background noise is removed from species audio data, and normalization and pre-emphasis are used to further process the audio to obtain the audio to be analyzed; species positioning data is converted into plane coordinates, and auxiliary data (such as data measured by an inertial measurement unit) is combined to obtain the positioning data to be analyzed. When preprocessing environmental data, for example, the environmental data is standardized to a unified range to obtain the environmental data to be analyzed.

[0050] After data preprocessing, image features, audio features, positioning features and environmental features are extracted from the preprocessed data to obtain multi-modal features, for example, image features are extracted using a convolutional neural network, mel-frequency spectrum features of audio are extracted, positioning features of positioning data are extracted, and environmental features of environmental data are extracted; the environmental features are obtained by converting the environmental data into a feature vector that can reflect the state of the living environment of the species, for example, the temperature, humidity and other environmental data are standardized as environmental features. When performing feature extraction, features related to species identification and spatial distribution need to be extracted; after feature extraction, multi-modal features are time and space aligned based on timestamps, which means that the spatial position information of biological data and environmental data needs to be integrated to ensure the consistency of data in time and space dimensions, providing a basis for subsequent population distribution analysis.

[0051] After the feature alignment is performed, the aligned image features, audio features, positioning features, and environmental features are combined with corresponding weights for feature splicing to generate a comprehensive feature vector. In generating the comprehensive feature vector, the aligned image features, audio features, positioning features, and environmental features can be first spliced for feature splicing to generate a first feature vector; and then initial weights are assigned to each modality according to the importance and correlation of different modal data, and the assigned initial weights are fused into the first feature vector to generate a second feature vector. The initial weights of the environmental features are determined based on the influence degree of the environmental data on the survival and identification of the species, and the initial weights of the modal features in the biological features are assigned based on the estimated importance of each modal feature in species identification and species spatial distribution quantification, for example. In the case of generating the second feature vector, the comprehensive feature vector is generated by dynamically adjusting the weights, and the process of dynamically adjusting the weights is based on the contribution of the biological features and the environmental features to the species identification and the species spatial distribution quantification. In dynamically adjusting the weights, the weights of the modal features can also be adjusted based on the attention mechanism. It should be noted that, since the embodiments of the present application need to quantify the spatial distribution of the species population, more attention needs to be paid to the features related to the species location when assigning the initial weights and dynamically adjusting the weights. In the implementation of the traditional method, multiple modal data (such as only image or audio) are relied on, which limits the recognition accuracy; at the same time, the traditional modal fusion lacks an alignment mechanism for cross-modal feature fusion, and there is a semantic gap problem; in addition, the weight allocation of each modality is mostly fixed weight allocation, which cannot adapt to the data quality differences in different scenarios. In the implementation process of the present scheme, cross-modal feature extraction is performed on biological data (image, audio, positioning) and environmental data (meteorology, natural resources), the data quality can be improved through data preprocessing, better data basis and stronger data support are provided for subsequent species identification; by aligning the multi-modal features in time and space, the consistency of the data in the time and space dimensions can be ensured; when the features are fused, the weights of the modal features and the features related to the species location are considered, and the comprehensive feature vector can be provided based on the species population spatial distribution quantification task and the species identification task.

[0052] The following will be described in conjunction with the accompanying drawings Figure 2 The specific implementation scheme of extracting features from biological data and environmental data and performing feature fusion based on the extracted features will be described by way of example.

[0053] A. Extracting basic features through a pre-trained multi-modal encoder. The bottom layer network parameters are shared across modalities, and the top layer generates standardized feature representations through a modality-specific adapter (Adapter).

[0054] B. Feature fusion with three-level contrastive learning constraints. Specifically, the following three ways are used for contrast: (1) instance-level contrast: ensure that the cross-modal samples of the same ecological event are adjacent in the embedding space; (2) feature-level contrast: minimize the Frobenius distance of different modal feature distributions; (3) semantic-level contrast: align high-level semantic concepts through CLIP loss. Hierarchical alignment is achieved through three-level contrastive learning (instance level, feature level, semantic level).

[0055] The overall scheme of extracting features and performing feature fusion is as follows:

[0056] I. Instance-level contrast, cross-modal positive and negative sample construction.

[0057] In the data preprocessing stage: for image data, the target area is detected using YOLOv5, and then center cropping is performed, finally the image resolution is adjusted to, for example, 224x224; for audio data, it is converted into a 128-dimensional Mel spectrogram through STFT (Short-Time Fourier Transform); for environmental data, it is normalized into, for example, a 10-dimensional feature vector.

[0058] In the sample pair generation stage: generate positive sample pairs, combine different modal data of the same spatiotemporal point, for example, pair image with audio, image with environmental data. Generate negative sample pairs, combine different modal data across spatiotemporal points at random, such as pairing image with audio at different time points, image at different time points with environmental data, etc.

[0059] In the loss calculation stage: use the contrastive loss function to calculate the ratio of the similarity between positive sample pairs and the similarity between negative sample pairs to optimize the model, so that the similarity of positive sample pairs is higher and the similarity of negative sample pairs is lower. Among them, the similarity measure adopts the dot product form of feature vectors s v , a = , T denotes the transpose operation of the matrix, W is a learnable projection matrix, and belong to the feature vector. For example, the initial temperature coefficient τ = 0.1, the temperature coefficient τ is a hyperparameter in the contrastive loss function, used to control the "temperature" of the similarity, i.e., the sensitivity of the similarity. Here τ is used to scale the similarity s v , a ), when τ is small, the degree of exponentialization of the similarity is high, which means that a small change in the similarity will result in a large change in the exponential term.

[0060] ​​Through instance-level contrast, the correlation between different modal data can be learned, the discrimination boundary strength of positive and negative sample pairs can be dynamically adjusted, and preliminary feature representations can be generated.

[0061] II. Feature-level contrast, modal distribution alignment.

[0062] Shared encoder architecture: a shared encoder is constructed, which includes shared convolutional layers (using pre-trained ResNet34) and adapters for different modalities. The adapters map the features of different modalities to a 256-dimensional shared space. Specifically, the adapter for the image modality maps 512-dimensional features to 256-dimensional features, the adapter for the audio modality maps 128-dimensional features to 256-dimensional features, and the adapter for the environment modality maps 10-dimensional features to 256-dimensional features.

[0063] Distribution alignment loss: the mean and covariance of the features of each modality are calculated, and the difference between the distributions of the two modalities is measured using, for example, the Frobenius norm and the Euclidean norm, to achieve distribution alignment. The statistical quantities are updated using a sliding average with a momentum coefficient β = 0.9.

[0064] Through feature-level contrast, the generated features have consistency in the shared space, the strictness of cross-modal distribution alignment is controlled, and the distribution difference between different modalities is reduced.

[0065] III. Semantic-level contrast, concept space mapping.

[0066] Label semantic expansion: an ecological knowledge graph is constructed to link species, habitats, and environmental factors, for example, species → habitat → environmental factor. The concept embedding vectors are extracted by encoding the species description text using, for example, a BERT model.

[0067] Loss function: the cosine similarity between visual features and concept embedding vectors is calculated, as well as the cosine similarity between auditory features and concept embedding vectors, and then the mean square error loss of the two similarities is returned, to achieve alignment of visual features and auditory features with semantic concepts.

[0068] Through semantic-level contrast, the clustering granularity of the concept space is balanced, and the generated features not only have consistency in distribution but also have consistency in semantic level.

[0069] During adjustment, for example, adaptive temperature control is performed in the following manner. In the three-level contrast learning framework, the adaptive temperature control mechanism does not directly act on the feature extraction process, but indirectly improves the quality of feature extraction by adjusting the feature representation optimization in the contrast learning stage.

[0070] 1. Confidence evaluation, confidence evaluation is the starting point of the whole dynamic temperature coefficient adjustment process. The feature quality is evaluated based on the feature norm, the norm of the feature vector is calculated, and normalization processing is performed to obtain the confidence.

[0071] 2. Temperature update rule, the temperature update rule is the core of the dynamic temperature coefficient adjustment, which dynamically adjusts the temperature coefficient according to the result of the confidence evaluation. Specifically: according to the average value of the confidence of the features in the batch, the basic temperature coefficient and the decay coefficient, the temperature coefficient is dynamically adjusted. For example, based on the basic temperature coefficient τ base = 0.15, the decay coefficient k = 0.05, dynamic adjustment is performed every batch (B = 256). Each batch (B = 256) means that each batch contains 256 samples.

[0072] 3. Hardware acceleration, hardware acceleration is the implementation means of dynamic temperature coefficient adjustment. When deploying using TensorRT, the temperature coefficient adjustment is implemented as a custom plugin layer to improve computing efficiency.

[0073] Through dynamic temperature coefficient adjustment, the scaling degree of similarity can be adaptively adjusted. This mechanism indirectly adjusts the optimization target of the feature encoder, which realizes: automatically reducing the influence of low-quality data on feature extraction, balancing the modal, and preventing a certain modal from dominating the feature representation space. The above embodiment constructs a framework from feature extraction to cross-modal fusion through a three-level adaptive temperature adjustment mechanism, effectively improves the hierarchical alignment mechanism of multi-modal pre-training, and reduces the gap between cross-modal semantics.

[0074] In the cross-modal feature fusion process, the adaptive temperature control mechanism adjusts the feature representation optimization of the contrast learning, which indirectly affects the modal weight distribution in the subsequent fusion stage; in the cross-modal fusion link, the features of different modalities are spliced, and the relationship between modalities is modeled through an attention mechanism, and finally fused through a weight matrix. The attention mechanism adopts, for example, dot product attention form, and the dimension of the weight matrix is determined according to the feature dimension. The embodiment of the present application adopts a dynamic weight adjustment mechanism, adaptively allocates the weight of each modal feature based on the feature contribution degree (such as confidence), and adapts to the data quality difference in different scenarios.

[0075] In the above process, instance-level comparison is the basis, the positive and negative sample pairs are constructed to strengthen the learning of the association between different modal data by the comprehensive feature vector generation model (which is a comprehensive multi-modal learning system for generating high-quality comprehensive feature vectors); feature-level comparison is based on instance-level comparison, further optimizes the consistency of features through feature distribution alignment; semantic-level comparison is based on feature-level comparison, further aligns the features of different modalities by introducing semantic information, and ensures that they have consistency at the semantic level; dynamic temperature coefficient adjustment runs through the entire training process, optimizes the training process of the model (comprehensive feature vector generation model) by dynamically adjusting the temperature coefficient, and improves the performance and generalization ability of the model; the connection with the comprehensive feature vector is the final goal, to ensure that the dynamic temperature coefficient adjustment mechanism is closely combined with the finally generated comprehensive feature vector, so as to generate high-quality comprehensive feature vectors for subsequent tasks.

[0076] As a specific application example, in the wetland bird monitoring scene, the input data includes: visual data (white egret flying image), audio data (200-800Hz calling spectrogram), and environmental data (air temperature 28℃, humidity 70%). The processing process includes: 1, instance-level comparison, strengthening the positive sample association of "white egret image-specific call"; 2, feature-level alignment, eliminating feature distribution offset caused by light changes; 3, semantic-level mapping, associating "bird-related-wetland environment" concepts. In the output comprehensive feature vector, for example, the first 256 dimensions are visual features: [0.12, -0.05,..., 0.78], the middle 256 dimensions are audio features: [0.33, 0.41,..., -0.12], and the last 256 dimensions are environmental features: [0.91, 0.02,..., 0.15].

[0077] In an optional embodiment of the present application, when species identification is performed based on a target large model, it includes:

[0078] Based on the target large model, the comprehensive feature vector is analyzed to obtain identification reference information of each species in the target ecological region, and the identification reference information includes at least one reference name and a confidence degree corresponding to the reference name;

[0079] For each species in the target ecological region, the identification reference information of the species is corrected based on the species prior information to obtain species identification information.

[0080] After obtaining the comprehensive feature vector, the target large model is optimized according to the comprehensive feature vector to enhance the understanding ability of the target large model to the multi-modal data. The optimization mode is, for example, to fuse the comprehensive feature vector with the feature representation of the target large model to enhance the understanding ability of the target large model to the multi-modal data. After optimizing the target large model, the optimized target large model is used to analyze the comprehensive feature vector and provide identification reference information of each species in the target ecological region. The identification reference information of the species includes at least one reference name and a confidence degree corresponding to the reference name.

[0081] After obtaining the identification reference information of the species, the identification reference information is corrected based on the species prior information of the species to obtain species identification information. The species prior information refers to information known or assumed about the characteristics, behavior, distribution, etc. of the species. The species prior information helps better understand and predict the behavior and distribution of the species. Correcting the identification reference information based on the species prior information is actually correcting the confidence degree corresponding to the reference name. After correcting the confidence degree corresponding to the reference name using the species prior information, the species identification information (including the species name and the confidence degree of the species name) is determined based on the confidence degree corresponding to the corrected at least one reference name, such as determining the reference name with the highest confidence degree as the final species name.

[0082] The above implementation process optimizes the understanding ability of the target large model, enhances the understanding ability of the target large model to the multi-modal data, and further improves the accuracy of model identification and the accuracy of cross-modal reasoning. Based on the optimized target large model, the comprehensive feature vector is processed, the identification reference information of the species is output, and the identification reference information is corrected based on the species prior information, which can relatively accurately identify the species and ensure the efficiency of species identification. The problems of poor adaptability of traditional models to multi-modal data, insufficient edge deployment capability, lack of prior knowledge correction link, and low confidence degree identification result prone to noise are effectively solved.

[0083] The process of quantifying the spatial distribution of the species population is introduced as follows. The species identification information includes the species name and the confidence degree of the species name. When quantifying the spatial distribution of the species population based on the species identification information corresponding to the target ecological region and the comprehensive feature vector, the species population quantification information is generated, including:

[0084] The species with a confidence degree of the species name in the target ecological region less than a preset threshold are filtered, and a species set is determined based on the remaining species. At least part of the species in the species set are key species, and the key species are species with an influence on the target ecological region meeting a preset condition.

[0085] Based on the species identification information and the comprehensive feature vectors corresponding to the key species in the species set, a population distribution map indicating the distribution position of the key species, a population density map indicating the density distribution of the key species, and a population dynamic change map indicating the change of the key species at different time points are generated.

[0086] After obtaining the species identification information corresponding to each species in the target ecological region, species filtering is performed based on the species name confidence included in the species identification information to filter out species with a species name confidence less than a preset threshold. Since the identification result with high confidence is generally more reliable, filtering out the identification result with low confidence can ensure that the species with high identification accuracy is provided.

[0087] After the species with a species name confidence less than the preset threshold is filtered out, a species set is determined based on the remaining species, at least part of the species in the species set being a key species. The key species can be understood as a species whose influence on the target ecological region meets a preset condition. For example, the species belonging to the key species meets at least one of the following conditions: 1. having an important role in maintaining the structure and function of the ecological system; 2. having a significant impact on the survival of other species and the ecological system; 3. having a unique role in the ecological system and being difficult to be replaced by other species.

[0088] After the identified species in the target ecological region are filtered to determine the species set, a population distribution map, a population density map, and a population dynamic change map are generated based on the species identification information and the comprehensive feature vectors corresponding to the key species in the species set. The generated population distribution map is used to indicate the position distribution of the key species in the geographical space, the generated population density map is used to indicate the density distribution of the key species in different regions, and the generated population dynamic change map is used to indicate the change of the key species at different time points.

[0089] It should be noted that when the application examples analyze the species identification information and the comprehensive feature vectors, quantify the spatial distribution of the species population, and generate the species population quantification information, a large model is not necessarily required to process it, and it relies more on GIS tools, statistical analysis methods, and visualization techniques. However, the large model can provide additional assistance in this process, such as predicting the future distribution and dynamic change of the species based on historical data and environmental factors; the large model can process complex multi-modal data to provide more in-depth analysis and insights; the large model can automatically process large-scale data to improve analysis efficiency. The large model processes time series and geographical spatial data in this stage, which can be used to predict the future distribution and dynamic change of the species, and provide in-depth analysis results; that is, the large model in this stage is different from the target large model used for species identification described above.

[0090] The method comprises the following steps:

[0091] extracting position information of the key species at a specified time from the comprehensive feature vector;

[0092] generating a population distribution map indicating the distribution of the key species at the specified time based on the position information of the key species at the specified time and the species identification information of the key species.

[0093] The population distribution map of the embodiment represents the distribution of the key species at a specified time (such as a certain time point or a certain time period). Before generating the population distribution map, the key species in the species set need to be identified. The generation of the population distribution map mainly involves the following operations: 1, geographical coordinate positioning; 2, species information association; 3, map drawing. When performing geographical coordinate positioning, the positioning information related to the observation position of the key species, such as longitude and latitude coordinates, is extracted from the comprehensive feature vector. These coordinate information comes from the positioning data provided by the collection device. When performing species information association, the species name of the key species is associated with the corresponding geographical coordinates. For example, if a key species is identified at a certain coordinate point, the species name of the key species is marked at the coordinate position. When drawing the map, the associated species name and coordinate point are drawn on the map with the geographical area as the background. For example, GIS software or related map drawing tools can be used to complete the drawing. Each species can be represented by different colors or symbols, so as to intuitively show the distribution position of different key species populations in space and form a population distribution map.

[0094] The method comprises the following steps:

[0095] dividing the target ecological region into a plurality of grids;

[0096] based on the position information of the key species at the specified time extracted from the comprehensive feature vector, distributing the key species into the corresponding grid;

[0097] for each grid, determining the grid weighted density based on the respective species weight of each key species contained in the grid, wherein the species weight is determined based on the species name confidence included in the species identification information corresponding to the key species;

[0098] determining the weighted density distribution of the target ecological region based on the respective grid weighted density of each grid, and visualizing the weighted density distribution of the target ecological region to generate a population density map.

[0099] The population density map in the embodiment represents the density distribution of the key species at a specified time. Before generating the population density map, the key species in the species set need to be identified. The generation of the population density map mainly involves the following stages: 1, regional division; 2, species statistics; 3, density calculation; 4, density map drawing. In the first stage, the target ecological region is divided into multiple grids. The size of the grid can be determined according to the accuracy requirement of the research and the density of the data. For example, in the area where the species distribution is relatively dense, smaller grids can be divided. In the second stage, based on the location information of the key species in the comprehensive feature vector at the specified time, the key species are assigned to the corresponding grid. In the third stage, the population density in each grid is calculated. When calculating the population density in the grid, the species weight corresponding to each key species contained in the grid also needs to be considered. The species weight can be determined based on the species name confidence, such as taking the species name confidence as the weight or taking the square of the species name confidence as the weight. After determining the species weight corresponding to each key species contained in the grid, the weights of the species are added to obtain the grid weighted density corresponding to the grid. When calculating, the number of each key species contained in the grid also needs to be considered. If there are multiple key species of the same kind in the grid, the species weight of each key species of the same kind needs to be added when calculating the grid weighted density. In the fourth stage, the weighted density distribution of the target ecological region is determined based on the grid weighted density corresponding to each grid in the target ecological region. The weighted density distribution is visualized to generate the population density map. For example, after calculating the grid weighted density corresponding to each grid, the population density map is drawn on the map using color gradient or contour line, etc. The darker the color or the denser the contour line, the higher the population density, and vice versa, so as to intuitively show the density of the key species population in different regions.

[0100] Since the confidence can be regarded as the probability of the actual existence of the species, the weighted operation based on the confidence is used to generate the weighted population density map, which can not only reflect the number of species, but also reflect the probability of the existence of the species, and at the same time make the density value in the map closer to the actual existence possibility of the species, so as to more truly reflect the distribution of the species in the geographical space; and the weighted population density map can more clearly highlight the areas with high species density, which helps to quickly identify the key ecological region.

[0101] The population density map generated above reflects the overall distribution of all key species in the grid, but cannot directly reflect the density of a single species. If it is necessary to analyze the density of a single species, the weighted density of a single key species in each grid can be calculated to obtain the weighted density distribution of the single key species in the entire target ecological region.

[0102] In the generation of the population dynamic change graph indicating the change of the key species at different time points based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, the method comprises the following steps of:

[0103] The target large model analyzes a plurality of time nodes of the comprehensive feature vector in the target time interval to generate a species identification time sequence, and the comprehensive feature vector is dynamically updated in the target time interval.

[0104] At each time node of the species identification time sequence, based on the comprehensive feature vector corresponding to the current time node and the species identification information corresponding to the key species in the species set, the number of species corresponding to each key species is counted, and / or the distribution of species corresponding to each key species is counted, to generate key species population information corresponding to the current time node.

[0105] The key species population information corresponding to each time node of the species identification time sequence is compared to generate a population dynamic change graph.

[0106] The target time interval in the embodiment is, for example, a historical time interval, a comprehensive time interval including a historical time interval and a future time interval, or a future time interval in special scenarios. Based on the plurality of time nodes of the comprehensive feature vector analyzed in the target time interval, the species identification time sequence is determined. The comprehensive feature vector is dynamically updated in the target time interval. For the case where the target time interval is associated with the future time interval, the comprehensive feature vector can be obtained by prediction, such as generating the comprehensive feature vector at a future time point based on historical data and a prediction model. Accordingly, since the species identification information is determined by the target large model analyzing the comprehensive feature vector, if the comprehensive feature vector is obtained by prediction, the matching species identification information can be obtained based on the target large model analyzing the predicted comprehensive feature vector. Generally, the species identification information obtained based on the target large model analyzing different comprehensive feature vectors has little difference, and the key species in the species set remains unchanged.

[0107] At each time node of the species identification time sequence, based on the comprehensive feature vector corresponding to the current time node and the species identification information, at least one of the number of species corresponding to each key species and the distribution of species corresponding to each key species is counted, and the key species population information corresponding to the current time node is generated based on the counting. For any time node, the comprehensive feature vector corresponding to the time node can be determined based on the collected data in the period corresponding to the time node and the previous time node; if the time node belongs to a future time node, the corresponding comprehensive feature vector can be obtained by prediction.

[0108] In generating the key species population information corresponding to the current time node, after determining the key species in the species set, at least one of the species quantity and location information of each key species is extracted from the comprehensive feature vector based on the species identification information of the key species; and the key species population information is generated according to at least one of the species quantity and species distribution of each key species, and the species distribution of the key species is determined based on the location information.

[0109] After determining the key species at the current time node, for each key species, the species quantity and / or location information of the key species is extracted from the comprehensive feature vector corresponding to the current time node. If the key species is an animal, the extracted location information is the species activity area information. After obtaining the location information of the key species, the species distribution of the key species can be determined. For any key species, after extracting the species quantity and / or location information, the statistical information corresponding to the key species can be determined, and the statistical information corresponding to each key species is aggregated to obtain the key species population information of the current time node.

[0110] After the key species population information of each time node is counted, the key species population information corresponding to each time node of the species identification time sequence is compared and visually displayed to generate a population dynamic change graph.

[0111] In another embodiment of generating a population dynamic change graph, the key species of a target ecological region is determined in advance, and the data related to the observation time of the key species is extracted from the multiple comprehensive feature vectors, and the observation records of the same key species at different time points are arranged into time sequence data. For example, the observation quantity and location information of a certain key species in different time intervals such as every day, every week, or every month is recorded. For each key species, the quantity change at different time points is analyzed, such as quantifying the dynamic change of the population quantity by calculating the quantity difference or change rate of adjacent time points. The location change of the key species at different time points is analyzed to determine the migration path or change of the activity range of the species. For example, by comparing the geographic coordinates at different times, it can be found that a certain key species migrates from one region to another region in a period of time. For multiple key species, the information of population quantity change and location change is combined and displayed on the map in a dynamic manner. For example, different colored lines are used to represent the migration path of the species, and the thickness of the line can represent the quantity scale of the migration; different sizes of circles are used to represent the species quantity at different time points, and the position of the circle represents the distribution position of the species, and with the passage of time, these graphical elements change on the map, thereby intuitively showing the dynamic change process of the key species population.

[0112] After determining the key species in the target ecological region, the above implementation process generates a population distribution map indicating the distribution location of the key species, a population density map indicating the density distribution of the key species, and a population dynamic change map indicating the change of the key species at different time points, to quantify the spatial distribution of the key species population. The scheme introduces a species weight calculation grid weighted density, which more truly reflects the ecological pressure and can facilitate researchers to understand the ecosystem structure of the target ecological region, determine the key area of biodiversity protection, and provide basic data for predicting the change of the ecosystem.

[0113] The scheme for generating a quantitative report is introduced as follows. Based on the species population quantification information and the collected raw data, a quantitative report describing the distribution characteristics of the species population is generated and displayed, including:

[0114] The population distribution map, the population density map, the population dynamic change map, and the collected raw data are comprehensively analyzed to generate statistical analysis results including at least the population number change, the population distribution range, the population density change, the hot spot area, the factors affecting the population dynamic change, and the ecological pressure of the target ecological region. Based on the statistical analysis results and the population distribution map, the population density map, and the population dynamic change map, a quantitative report is generated and displayed.

[0115] The raw data provided by the collection device can provide more detailed information for verifying the accuracy of the quantitative results, and the detailed background information provided by the raw data can also help to understand the reasons for the species distribution. The generated population distribution map is used to show the distribution of the species in the geographical space, the generated population density map is used to show the density distribution of the species in different regions, and the generated population dynamic change map is used to show the distribution and number change of the species at different time points.

[0116] By combining the population distribution map, the population density map, the population dynamic change map, and the raw data, the following analysis can be performed:

[0117] The number of each key species in each time period is counted, and the change trend is analyzed to obtain the population number change; the distribution range of the species in the geographical space is analyzed to determine the activity area of the species to obtain the population distribution range; the density distribution of the species in different areas and the change thereof are calculated and visualized to obtain the population density change; the area with high density of the species is identified to determine the hotspot area of the species; the influence of environmental data on the population dynamic change is analyzed to obtain the factors influencing the population dynamic change; and the ecological pressure of the target ecological area is evaluated, including the comprehensive influence of the number, distribution range, density change of the species and environmental factors. It should be noted that when analyzing in combination with the population distribution map, the population density map, the population dynamic change map and the original data, multiple population distribution maps and multiple population density maps can be combined to compare the population distribution and the population density at different times. The multiple population distribution maps and the multiple population density maps can be obtained based on the adaptive multiple comprehensive feature vectors.

[0118] The statistical analysis results obtained based on the above analysis include: population number change showing the change of the number of species over time, population distribution range showing the distribution range of the species in the geographical space, population density change showing the density distribution of the species in different areas and the change thereof, hotspot area with high density of the species, factors influencing the population dynamic change (environmental factors), and ecological pressure of the target ecological area.

[0119] After obtaining the statistical analysis results, the statistical analysis results, the population distribution map, the population density map and the population dynamic change map are summarized to automatically generate and display a quantitative report containing statistical analysis (such as migration path and environmental impact), thereby providing high-quality quantitative reports for researchers and providing scientific basis for ecological protection and management.

[0120] The complex ecological system key species population spatial distribution quantification method provided by the embodiments of the present application will be introduced below through an overall implementation process, as shown in Figure 3 The method includes the following steps:

[0121] Step 301, preprocessing biological data and environmental data collected in a target ecological area.

[0122] Step 302, feature extraction is performed in the preprocessed data to obtain multi-modal features, and the multi-modal features are time and space aligned. The multi-modal features include image features, audio features, positioning features and environmental features.

[0123] Step 303, the image features, audio features, positioning features and environmental features subjected to the alignment processing are combined with corresponding weights for feature splicing to generate a comprehensive feature vector.

[0124] Step 304, based on the target large model, analyzing the comprehensive feature vector to obtain identification reference information of each species in the target ecological region, the identification reference information including at least one reference name and a confidence degree corresponding to the reference name.

[0125] Step 305, for each species in the target ecological region, based on the species prior information, correcting the identification reference information of the species to obtain species identification information, the species identification information including a species name and a species name confidence degree.

[0126] Step 306, filtering the species whose species name confidence degree is less than a preset threshold in the target ecological region, determining a species set based on the reserved species, at least part of the species in the species set belonging to key species.

[0127] Step 307, based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, generating a population distribution map indicating the distribution position of the key species, a population density map indicating the density distribution of the key species, and a population dynamic change map indicating the change of the key species at different time points.

[0128] Step 308, comprehensively analyzing the population distribution map, the population density map, the population dynamic change map, and the collected original data to generate statistical analysis results including at least population quantity change, population distribution range, population density change, hot spot area, factors affecting population dynamic change, and ecological pressure of the target ecological region.

[0129] Step 309, based on the statistical analysis results and the population distribution map, the population density map, and the population dynamic change map, generating and displaying a quantitative report.

[0130] In the overall implementation process shown in Figure 3 , the large model technology is used to realize intelligent and efficient species identification, and the multi-modal feature fusion is used to provide comprehensive feature representation, which can improve the accuracy of species identification; by combining the species identification information and the comprehensive feature vector to analyze and generate species population quantitative information, and combining the species population quantitative information and the original data to generate a quantitative report, the distribution and dynamic change of the species can be intuitively displayed, and a comprehensive and accurate high-quality quantitative report can be provided, which provides a strong basis for ecological protection and management.

[0131] The embodiment of the present application provides a key species population spatial distribution quantification system in a complex ecological system, as shown in Figure 4 , comprising:

[0132] The extraction generation module 401 is configured to perform feature extraction on biological data and environmental data collected in a target ecological region, and fuse the extracted multi-modal features to generate a comprehensive feature vector. The biological data at least includes species image data, species audio data, and species positioning data. The environmental data at least includes meteorological data and natural resource data.

[0133] The analysis and identification module 402 is configured to analyze the comprehensive feature vector based on a target large model, identify species in the target ecological region, and obtain species identification information of each species in the target ecological region.

[0134] The quantification generation module 403 is configured to quantify the spatial distribution of species population based on the species identification information corresponding to the target ecological region and the comprehensive feature vector, and generate species population quantification information. The species population quantification information at least includes a population distribution map, a population density map, and a population dynamic change map.

[0135] The generation and display module 404 is configured to generate a quantification report that comprehensively describes the distribution characteristics of the species population based on the species population quantification information and the collected original data, and display the quantification report.

[0136] Optionally, the extraction generation module includes:

[0137] The preprocessing submodule is configured to perform data preprocessing on the collected biological data and environmental data.

[0138] The extraction submodule is configured to perform feature extraction on the preprocessed data to obtain multi-modal features, including biological features and environmental features. The biological features include image features, audio features, and positioning features.

[0139] The alignment submodule is configured to perform time and space alignment on the multi-modal features.

[0140] The splicing generation submodule is configured to combine the image features, audio features, positioning features, and environmental features that have undergone alignment processing with corresponding weights to perform feature splicing, and generate the comprehensive feature vector.

[0141] The weights corresponding to each modal feature are dynamically adjusted based on feature contribution.

[0142] Optionally, the analysis and identification module includes:

[0143] The analysis and acquisition submodule is configured to analyze the comprehensive feature vector based on the target large model, and obtain identification reference information of each species in the target ecological region. The identification reference information includes at least one reference name and a confidence degree corresponding to the reference name.

[0144] The correction and acquisition submodule is used to correct the species identification reference information for each species in the target ecological area based on the species prior information to obtain the species identification information.

[0145] Optionally, the species identification information includes the species name and the species name confidence; and the quantification generation module includes:

[0146] A filtering and determining submodule, configured to filter species within the target ecological area whose species name confidence is less than a preset threshold, and determine a species set based on the retained species;

[0147] a generation submodule for generating, based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, a population distribution map indicating the distribution location of the key species, a population density map indicating the density distribution of the key species, and a population dynamic change map indicating the change of the key species at different time points;

[0148] Among them, at least some of the species in the species set belong to the key species, and the key species are species whose influence on the target ecological area meets preset conditions.

[0149] Optionally, the generating submodule includes:

[0150] An extraction unit, configured to extract the location information of the key species at a specified time from the comprehensive feature vector;

[0151] A generating unit is configured to generate a population distribution map indicating the distribution of the key species at the specified time based on the location information of the key species at the specified time and the species identification information of the key species.

[0152] Optionally, the generating submodule includes:

[0153] A division unit, used for dividing the target ecological area into a plurality of grids;

[0154] an allocating unit, configured to allocate the key species to corresponding grids based on the location information of the key species at a specified time extracted from the comprehensive feature vector;

[0155] a determining unit configured to determine, for each grid, a grid weighted density based on species weights corresponding to respective key species contained in the grid, wherein the species weights are determined based on confidence levels of species names contained in species identification information corresponding to the key species;

[0156] A determination generating unit is configured to determine a weighted density distribution of the target ecological region based on the grid weighted densities corresponding to each grid, and to visualize the weighted density distribution of the target ecological region to generate the population density map.

[0157] Optionally, the generating submodule comprises:

[0158] a record generating unit configured to record a plurality of time nodes at which the target large model analyzes the comprehensive feature vector within a target time interval, and generate a species identification time sequence, the comprehensive feature vector being dynamically updated within the target time interval;

[0159] a statistics generating unit configured to, at each time node of the species identification time sequence, based on the comprehensive feature vector corresponding to the current time node and the species identification information corresponding to the key species in the species set, count the number of species corresponding to each key species and / or count the distribution of species corresponding to each key species, and generate key species population information corresponding to the current time node;

[0160] a comparison generating unit configured to compare the key species population information corresponding to each time node of the species identification time sequence, and generate the population dynamic change graph.

[0161] Optionally, the statistics generating unit is further configured to:

[0162] after determining the key species in the species set, extract at least one of the number of species and the position information of each key species in the comprehensive feature vector based on the species identification information of the key species;

[0163] generate the key species population information according to at least one of the number of species and the distribution of species of each key species, the distribution of species of the key species being determined based on the position information.

[0164] Optionally, the generating and displaying module comprises:

[0165] an analysis generating submodule configured to comprehensively analyze the population distribution graph, the population density graph, the population dynamic change graph, and the collected original data, and generate statistical analysis results including at least population number change, population distribution range, population density change, hot spot area, factors affecting population dynamic change, and ecological pressure of the target ecological region;

[0166] a generating and displaying submodule configured to generate the quantitative report and display based on the statistical analysis results and the population distribution graph, the population density graph, and the population dynamic change graph.

[0167] For the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts are described in the part of the method embodiment.

[0168] The embodiment of the application further provides an electronic device, comprising: a processor, a memory, a computer program stored in the memory and executable on the processor, the computer program, when executed by the processor, implements each process of the method for quantifying the spatial distribution of the population of the key species in the complex ecosystem and achieves the same technical effects. To avoid repetition, details are not described herein.

[0169] For example, Figure 5 An entity structure diagram of an electronic device is shown. As Figure 5 shown, the electronic device can include a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 complete mutual communication through the communications bus 540. The processor 510 can invoke a logical instruction in the memory 530, and the processor 510 is configured to execute each process of the method for quantifying the spatial distribution of the population of the key species in the complex ecosystem, which is not described herein.

[0170] In addition, the logical instruction in the memory 530 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the application can be embodied in the form of a software product in essence or in the form of a part of the existing technology or part of the technical solutions, and the computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the application.

[0171] The embodiment of the application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program, when executed by a processor, implements each process of the method for quantifying the spatial distribution of the population of the key species in the complex ecosystem and achieves the same technical effects. To avoid repetition, details are not described herein. The computer readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0172] It should be noted that, in the present document, the terms "comprises / comprising" or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, and of course, they can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product in essence or in the form of a part of the prior art that makes a contribution. The computer software product is stored in a storage medium (such as a ROM / RAM, a magnetic disk, or an optical disk) and includes a plurality of instructions for causing a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods described in the various embodiments of the present application.

[0174] The embodiments of the present application are described above in combination with the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative and not restrictive. Those skilled in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the scope of protection of the claims.

Claims

1. A method for quantifying the spatial distribution of key species populations in complex ecosystems, characterized by: include: Extracting features from biological data and environmental data collected in the target ecological area, and fusing the extracted multimodal features to generate a comprehensive feature vector, wherein the biological data includes at least species image data, species audio data, and species location data, and the environmental data includes at least meteorological data and natural resource data; Analyzing the comprehensive feature vector based on the target large model, identifying species in the target ecological area, and obtaining species identification information of each species in the target ecological area; Based on the species identification information corresponding to the target ecological region and the comprehensive feature vector, the spatial distribution of species populations is quantified to generate species population quantitative information, including: Based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, a population distribution map indicating the distribution location of the key species, a population density map indicating the density distribution of the key species, and a population dynamic change map indicating the change of the key species at different time points are generated, wherein generating the population density map indicating the density distribution of the key species includes: Dividing the target ecological area into a plurality of grids; Based on the location information of the key species at a specified time extracted from the comprehensive feature vector, the key species are allocated to corresponding grids; For each grid, the grid weighted density is determined based on the species weights corresponding to the key species contained in the grid, wherein the species weights are determined based on the confidence level of the species name contained in the species identification information corresponding to the key species; Determining a weighted density distribution of the target ecological region based on the grid weighted densities corresponding to each grid, and visualizing the weighted density distribution of the target ecological region to generate the population density map; The species population quantitative information includes at least a population distribution map, a population density map, and a population dynamic change map; Based on the quantitative information of the species population and the collected original data, a quantitative report that comprehensively describes the distribution characteristics of the species population is generated and displayed.

2. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to claim 1, characterized in that: The feature extraction of biological data and environmental data collected in the target ecological area and the fusion of the extracted multimodal features to generate a comprehensive feature vector include: Perform data preprocessing on the collected biological data and environmental data; Extracting features from the preprocessed data to obtain multimodal features, wherein the multimodal features include biological features and environmental features, and the biological features include image features, audio features, and positioning features; aligning the multimodal features in time and space; The aligned image features, audio features, positioning features, and environmental features are combined with corresponding weights to perform feature splicing to generate the comprehensive feature vector; Among them, the weight corresponding to each modal feature is dynamically adjusted based on the feature contribution.

3. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to claim 2, characterized in that: The step of analyzing the comprehensive feature vector based on the target large model, identifying species within the target ecological area, and obtaining species identification information of each species within the target ecological area includes: Analyzing the comprehensive feature vector based on the target macromodel to obtain identification reference information of each species in the target ecological area, wherein the identification reference information includes at least one reference name and a confidence level corresponding to the reference name; For each species in the target ecological area, the species identification reference information is corrected based on the species prior information to obtain the species identification information.

4. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to any one of claims 1 to 3, characterized in that: The species identification information includes the species name and the confidence level of the species name; Filtering species within the target ecological area whose species name confidence is less than a preset threshold, and determining the species set based on the retained species; Among them, at least some of the species in the species set belong to the key species, and the key species are species whose influence on the target ecological area meets preset conditions.

5. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to claim 4, characterized in that: Based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, a population distribution map indicating the distribution location of the key species is generated, including: extracting the location information of the key species at a specified time from the comprehensive feature vector; Based on the location information of the key species at the specified time and the species identification information of the key species, a population distribution map indicating the distribution of the key species at the specified time is generated.

6. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to claim 4, characterized in that: Based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, a population dynamic change diagram indicating the change of the key species at different time points is generated, including: Recording multiple time nodes of the target large model analyzing the comprehensive feature vector within a target time interval to generate a species identification time series, wherein the comprehensive feature vector is dynamically updated within the target time interval; At each time node of the species identification time series, based on the comprehensive feature vector corresponding to the current time node and the species identification information corresponding to the key species in the species set, the number of species corresponding to each key species is counted, and / or the species distribution corresponding to each key species is counted to generate the key species population information corresponding to the current time node; The key species population information corresponding to each time node of the species identification time series is compared to generate the population dynamic change diagram.

7. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to claim 6, characterized in that: The method of generating key species population information corresponding to the current time node by counting the number of species corresponding to each key species and / or counting the species distribution corresponding to each key species based on the comprehensive feature vector corresponding to the current time node and the species identification information corresponding to the key species in the species set includes: After determining the key species in the species set, extracting at least one of species quantity and location information of each key species from the comprehensive feature vector based on the species identification information of the key species; The key species population information is generated based on at least one of the species quantity and species distribution of each key species, and the species distribution of the key species is determined based on the location information.

8. The method for quantifying the spatial distribution of key species populations in complex ecosystems according to claim 1, characterized in that: Based on the quantitative information of the species population and the collected original data, a quantitative report that comprehensively describes the distribution characteristics of the species population is generated and displayed, including: Conduct a comprehensive analysis of the population distribution map, the population density map, the population dynamics map, and the collected raw data to generate statistical analysis results that include at least population size changes, population distribution ranges, population density changes, hotspot areas, factors affecting population dynamics, and ecological pressures in the target ecological region; Based on the statistical analysis results and the population distribution map, the population density map, and the population dynamic change map, the quantitative report is generated and displayed.

9. A system for quantifying the spatial distribution of key species populations in complex ecosystems, characterized by: include: An extraction and generation module for extracting features from biological data and environmental data collected in the target ecological area, and fusing the extracted multimodal features to generate a comprehensive feature vector, wherein the biological data includes at least species image data, species audio data, and species location data, and the environmental data includes at least meteorological data and natural resource data; An analysis and identification module, configured to analyze the comprehensive feature vector based on the target large model, identify species within the target ecological area, and obtain species identification information of each species within the target ecological area; A quantitative generation module is used to quantify the spatial distribution of species populations based on the species identification information corresponding to the target ecological region and the comprehensive feature vector, and generate species population quantitative information, which is specifically used to: Based on the species identification information corresponding to the key species in the species set and the comprehensive feature vector, a population distribution map indicating the distribution location of the key species, a population density map indicating the density distribution of the key species, and a population dynamic change map indicating the changes of the key species at different time points are generated; it is also specifically used to generate the population density map indicating the density distribution of the key species, including: Dividing the target ecological area into a plurality of grids; Based on the location information of the key species at a specified time extracted from the comprehensive feature vector, the key species are allocated to corresponding grids; For each grid, the grid weighted density is determined based on the species weights corresponding to the key species contained in the grid, wherein the species weights are determined based on the confidence level of the species name contained in the species identification information corresponding to the key species; Determining a weighted density distribution of the target ecological region based on the grid weighted densities corresponding to each grid, and visualizing the weighted density distribution of the target ecological region to generate the population density map; The species population quantitative information includes at least a population distribution map, a population density map, and a population dynamic change map; A generation and display module is used to generate and display a quantitative report that comprehensively describes the distribution characteristics of species populations based on the quantitative information of the species populations and the collected original data.

Images