CN-121999354-A - Alga community in-situ intelligent identification method and system

CN121999354ACN 121999354 ACN121999354 ACN 121999354ACN-121999354-A

Abstract

The invention relates to the technical field of water ecology monitoring, in particular to an in-situ intelligent identification method and system for an alga community. Based on nonlinear mapping of a dynamic system and knowledge-driven control theory, an environment-feature dynamic mapping reference model and a biological constraint rule set are constructed. And performing multi-level quality self-checking through synchronous acquisition of multi-mode characteristics and environmental parameters. And (3) adopting a double-circulation dynamic correction mechanism, carrying out interpolation and re-correction on the characteristics by combining an internal circulation with a biological rule, and verifying the logical self-consistency among modes by an external circulation to generate a high-confidence reference characteristic model. And calculating fusion confidence through weighting fusion multi-mode similarity, and executing layering decision. The recognition result is constructed as an environment-feature-category triplet knowledge base, the increment of the asynchronous driving model and the rule set is updated, and the adaptability and recognition accuracy of the system are continuously improved. The method significantly enhances the dynamic response capability and long-term stability of algae monitoring.

Inventors

XU DAWEI
ZHANG JINLIANG
LI SUFANG
ZHENG ZHENRONG
HU YANPING
LI CHUNLIN
XU HONGWEI
LIU LIN
HUANG YANJI
ZHANG MINGJIE
ZHANG YANXIA
YAO WEI

Assignees

杭州腾海科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260409

Claims (10)

1. An in-situ intelligent identification method for an algae community is characterized by comprising the following steps: Constructing an environment-feature dynamic mapping reference model and a biological constraint rule set; synchronously capturing multi-modal feature data and real-time environment parameters of an algae community in a target water area, executing quality self-checking, and outputting qualified data of the multi-modal feature data and the real-time environment parameters only when the quality self-checking is completely passed; Based on real-time environmental parameters in the qualified data, calling a benchmark model to dynamically correct algae benchmark characteristics to generate an instantaneous reference characteristic model, and executing double-cycle verification, and outputting the high-confidence instantaneous reference characteristic model only when the double-cycle verification passes; matching and identifying the algae multi-modal features in the qualified data with the high-confidence instantaneous reference feature model, calculating the fusion confidence and executing layering decision; and constructing an environment-feature-category triplet knowledge base by the identification result and the corresponding high-confidence multi-modal features, environment parameters and correction parameters, and triggering asynchronous incremental updating of the reference model and the rule set.
2. The method for intelligently identifying algae communities in situ according to claim 1, wherein the constructing of the environment-feature dynamic mapping reference model comprises the following steps: Extracting a training sample set containing environment parameter vectors and algae multi-modal feature vectors; Constructing a deep neural network model, setting the number of nodes of an input layer of the network to be equal to the dimension of environmental parameters, setting the number of nodes of an output layer to be equal to the dimension of multi-mode characteristics, and adopting a multi-layer full-connection structure and a nonlinear activation function for a hidden layer; Defining a nonlinear mapping function from an environmental parameter space to a multi-modal feature space; Defining a loss function in a mean square error form to measure the difference between the predicted characteristic and the real characteristic; and iteratively updating network parameters by adopting a random gradient descent optimization algorithm, monitoring the degree of fitting through a verification set, and storing the network parameters with optimal generalization performance.
3. The method of claim 2, wherein the constructing the biological constraint rule set comprises: Extracting a reasonable value range of the algae feature vector, and defining the reasonable value range as a section set formed by the minimum value and the maximum value of the feature component; Extracting a proportion constraint relation between characteristic components, and defining a constraint form that the ratio is positioned in a preset interval; Extracting a gradient constraint rule of the feature changing along with the environmental parameter, wherein the gradient constraint rule is defined as that the absolute value of the partial derivative of the feature on the environmental parameter does not exceed a preset maximum value; formally encoding the constraint conditions to generate a structured rule set.
4. The method for intelligently identifying algae communities in situ according to claim 1, wherein the quality self-inspection comprises environmental parameter coverage verification, and the environmental parameter coverage verification method is as follows: Loading an effective coverage area defined by an environment parameter convex hull of the training sample set; judging whether the real-time environment parameter vector is positioned in the effective coverage area or not; if the data are located outside, judging that the environment is abnormal, triggering an environment anchor point data supplementing process, searching anchor point data closest to the real-time environment parameters and corresponding theoretical features as temporary reference standard, and instructing the in-situ monitoring equipment to perform repeated sampling for a plurality of times to obtain statistical features so as to enhance the adaptability of the edge environment.
5. The method for intelligently identifying algae communities in situ according to claim 4, wherein the quality self-inspection further comprises a mode characteristic data quality inspection and a mode space-time consistency inspection; The quality inspection of the modal characteristic data is that the microscopic image modal, the spectrum modal and the flow cell modal data are respectively and independently subjected to quality inspection, if any modal data is unqualified, the re-acquisition process is triggered only for the unqualified mode; The inter-mode space-time consistency test is to check the time consistency and the space consistency among different mode data on the basis that the quality of each single mode data is qualified, and if the check is abnormal, the target area directional re-scanning is triggered to compensate the deviation.
6. The method of claim 1, wherein the dual cycle verification comprises performing an inner cycle verification, the inner cycle verification performing step comprising: Acquiring an instantaneous reference feature model of the current iteration, and calling a biological constraint rule set to carry out item-by-item constraint verification; If the characteristic components which do not meet the constraint conditions exist, searching a plurality of anchor point data points closest to the real-time environment parameters in an environment anchor point database, and calculating a new reference characteristic model by adopting an inverse distance weighted interpolation method; updating a circulation counter, and circularly executing verification and recalibration until the maximum circulation times are passed through verification or reached; the dual cycle verification further includes executing an outer cycle verification, the outer cycle verification executing step including: The method comprises the steps of obtaining an instantaneous reference characteristic model passing through internal circulation verification, performing inter-mode biological logic self-consistency verification, and calculating a comprehensive self-consistency score, wherein a calculation formula is as follows: , wherein, In order to integrate the self-consistency score, For the number of self-consistency checks, Is the first A score for the individual self-consistency check; if the score is lower than a preset threshold, analyzing the feature components which are not self-consistent and tracing back to an internal circulation stage, adjusting a weight distribution strategy in interpolation correction or introducing new constraint conditions, and re-executing an internal circulation interpolation re-correction and verification process.
7. The method for intelligently identifying algae communities in situ according to claim 1, wherein the matching identification comprises: Extracting reference features in an actually collected algae multi-modal feature and high-confidence instantaneous reference feature model; calculating a structural similarity index of a microscopic image mode, calculating a spectrum angle matching cosine value of a spectrum mode, and calculating a Papanicolaou distance of a flow cell mode; Setting weight coefficients of all modes, and carrying out weighted fusion on the similarity of all modes to calculate fusion confidence; the structural similarity index of the microscopic image mode is obtained by comparing the mean value, the variance and the covariance of the two images, the spectrum angle matching cosine value of the spectrum mode is obtained by calculating the included angle cosine value between the two spectrum vectors, and the Pasteur distance of the flow cytometry mode is obtained by calculating the Pasteur distance of the two cell groups in the characteristic space.
8. The method for intelligently identifying algae communities in situ according to claim 7, wherein the hierarchical decision comprises: Setting a low confidence threshold and a high confidence threshold; If the fusion confidence coefficient is lower than the low confidence coefficient threshold value, retransmitting the current data to a dynamic correction stage to carry out feature correction and verification again; If the fusion confidence coefficient is between the low confidence coefficient threshold value and the high confidence coefficient threshold value, triggering a microscale rescanning process of the target area to acquire finer data; If the fusion confidence is higher than Gao Zhixin degrees threshold, directly outputting the recognition result.
9. The method of claim 1, wherein the triggering asynchronous incremental updates of the benchmark model and the rule set comprises: Monitoring the number of newly added environment-feature-category triples, and triggering a background increment updating task when the number reaches a preset threshold; Analyzing the environmental parameter distribution of the incremental sample, identifying points outside the original coverage area as candidate anchor points, generating new anchor points after cluster analysis, and expanding the coverage area; And analyzing the feature distribution and constraint violation conditions of the increment sample, mining the new proportional relation between the feature boundary and the feature, and updating the constraint condition of the rule set.
10. An intelligent in situ identification system for algal communities according to any one of claims 1-9, wherein the system comprises: the model construction module is used for constructing an environment-feature dynamic mapping reference model and a biological constraint rule set; The data acquisition and verification module is used for synchronously capturing multi-modal characteristic data and real-time environment parameters of the algae community in the target water area, executing quality self-checking, and outputting qualified data of the multi-modal characteristic data and the real-time environment parameters only when the quality self-checking completely passes; The dynamic correction module is used for calling the standard model to dynamically correct the algae standard feature based on the real-time environment parameters in the qualified data to generate an instantaneous reference feature model, executing double-cycle verification, and outputting the high-confidence instantaneous reference feature model only when the double-cycle verification passes; the identification decision module is used for carrying out matching identification on the algae multi-modal characteristics in the qualified data and the high-confidence instantaneous reference characteristic model, calculating the fusion confidence and executing layering decision; The knowledge updating module is used for constructing an environment-feature-category triplet knowledge base by using the identification result and the corresponding high-confidence multi-mode features, environment parameters and correction parameters, and triggering asynchronous increment updating of the reference model and the rule set.

Description

Alga community in-situ intelligent identification method and system Technical Field The invention relates to the technical field of water ecology monitoring, in particular to an in-situ intelligent identification method and system for an alga community. Background The algae community is an important indicator organism of the health condition of the water ecological system, and the variety composition and abundance change of the algae community directly reflect the nutrition state and eutrophication risk of the water body. The traditional algae identification mainly depends on manual sampling and laboratory microscopic examination modes, and the method is time-consuming, labor-consuming, low in efficiency, difficult to realize real-time monitoring, serious in hysteresis of monitoring results, and incapable of providing timely data support for early warning and emergency treatment of water bloom. In addition, manual microscopic examination has extremely high requirements on professional knowledge of operators, has strong subjectivity, can have large difference in recognition results among different operators, and is difficult to ensure consistency and comparability of data. With the development of image processing and artificial intelligence technology, an automatic algae identification method based on microscopic images gradually becomes a research hotspot. In the prior art, characteristics such as morphology, texture, pigment and the like of algae are extracted by collecting images of the algae, and species identification is carried out by utilizing a pre-trained classification model. However, microscopic image-based automatic algae identification methods are mostly model trained based on standard data sets under static, laboratory conditions. In an actual in-situ monitoring environment, the dynamic changes of environmental parameters such as water temperature, illumination, nutrient salt concentration and the like can obviously influence the morphological characteristics, the size, the pigment content and the like of algae cells, namely, the phenotype plasticity exists. When the real-time environment parameters deviate from the coverage range of the training data, the recognition accuracy of the static model can be drastically reduced, and erroneous judgment or missed judgment can be caused. Furthermore, the prior art has obvious disadvantages in terms of data quality control. The in-situ monitoring equipment is placed under water for a long time, and is easily influenced by factors such as biofouling, bubble interference, illumination change and the like, so that the acquired image quality is uneven. Low quality data input can directly affect the accuracy of subsequent identification. Although partial schemes introduce data cleaning steps, most adopt single threshold judgment, lack of depth consideration on environment-feature coupling relation, and lack of effective verification mechanism on space-time consistency among different modal features, so that robustness of a model to abnormal data is poor, and long-term, stable and high-precision in-situ monitoring requirements are difficult to meet. Disclosure of Invention In order to solve the problems of difficult adaptation to dynamic environment change, difficult in-situ data quality assurance, low recognition accuracy and poor robustness in the prior art, the invention provides an alga community in-situ intelligent recognition method and system, which are characterized in that an environment-feature dynamic mapping reference model is constructed and biological rule constraint is assisted, by combining the technical scheme of multi-level quality self-checking, double-circulation dynamic correction and verification, layered decision output and knowledge base increment updating mechanism, the intelligent recognition of high confidence and high adaptability to the algae community in the dynamic water area environment is realized. The technical scheme of the application specifically comprises the following steps: according to an aspect of the present application, there is provided an in-situ intelligent identification method of an algal colony, comprising: Constructing an environment-feature dynamic mapping reference model and a biological constraint rule set; synchronously capturing multi-modal feature data and real-time environment parameters of an algae community in a target water area, executing quality self-checking, and outputting qualified data of the multi-modal feature data and the real-time environment parameters only when the quality self-checking is completely passed; Based on real-time environmental parameters in the qualified data, calling a benchmark model to dynamically correct algae benchmark characteristics to generate an instantaneous reference characteristic model, and executing double-cycle verification, and outputting the high-confidence instantaneous reference characteristic model only when the double-cycle verification passes; matching