Water Quality Monitoring Method and System Based on AI Optical Sensor

Abstract

The present application provides a water quality monitoring method and system based on an AI optical sensor, which relates to the field of water quality monitoring. By monitoring water body parameters, it tracks the time-varying water body environmental conditions that cause spectral signals to be interfered. In addition, by establishing a dynamic mapping between spectral features and environmental time series features, spatio-temporal decoupling of interfering components is achieved. That is to say, by learning the mapping relationship between the time series features of water body parameters and spectral interference, environmental interference analysis is carried out in the way of feature matching, thus solving the problem that the influence of environmental parameters on the spectrum has a time delay, so as to more accurately predict the spectral interference at the current moment, and eliminate or weaken the interference of water body environmental factors on the spectral signal from the original spectral data, thereby obtaining purer water body component spectral feature information and water quality category information.

Description

Technical Field

[0001] This application relates to the field of water quality monitoring, and more particularly, in the embodiments of this application, it relates to a water quality monitoring method and system based on an AI optical sensor. Background Art

[0002] With the increasingly severe global water environment problems, water quality monitoring technology is developing rapidly towards the direction of intelligence and real-time. Although the traditional chemical analysis method has high accuracy, it has limitations such as long time consumption, large reagent consumption, and difficulty in realizing in-situ monitoring. The water quality monitoring technology based on optical sensors has become a research hotspot in the field of environmental monitoring due to its advantages such as non-contact, rapid response, and continuous monitoring.

[0003] However, the spectral signals collected by optical sensors are the result of the combined action of water body components and environmental factors. Specifically, the dynamic changes of environmental factors (such as temperature changes, turbidity fluctuations, etc.) will produce coupling interference with the spectral characteristics of target components, resulting in the phenomena of "same substance, different spectra" and "different substances, same spectra". That is to say, this interference will change the baseline drift, scattering effect, absorption intensity, etc. of the spectrum, thus masking or distorting the true spectral information of the water body components, resulting in errors when the model conducts water quality monitoring based on the interfered spectral data and reducing the monitoring accuracy.

[0004] Therefore, an optimized water quality monitoring scheme based on optical sensors is desired. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide a water quality monitoring method and system based on an AI optical sensor, which monitor water body parameters to track the time-varying water body environmental conditions that cause spectral signals to be interfered. In addition, by establishing a dynamic mapping between spectral characteristics and environmental time series characteristics, the spatio-temporal decoupling of interfering components is achieved. That is to say, by learning the mapping relationship between the time series characteristics of water body parameters and spectral interference, environmental interference analysis is carried out in the way of feature matching, thus solving the problem that the influence of environmental parameters on the spectrum has a time delay, so as to more accurately predict the spectral interference at the current moment, and eliminate or weaken the interference of water body environmental factors on the spectral signal from the original spectral data, thereby obtaining purer spectral characteristic information of water body components and water quality category information.

[0006] According to one aspect of this application, a water quality monitoring method based on an AI optical sensor is provided, which includes:

[0007] Collecting spectral map data of the monitored water body through an optical sensor;

[0008] Real-time monitoring of the water temperature data and water turbidity data of the monitored water body through a sensor group;

[0009] Based on the real-time monitored water temperature data and water turbidity data of the monitored water body, a sequence of water body temperature time series characteristics and a sequence of water body turbidity time series characteristics are obtained;

[0010] Extracting spectral semantic features of the water body based on the spectral data of the monitored water body;

[0011] The water body spectral semantic features are used as reference information matching features, and a feature sequence composed of a sequence of water body temperature time series features and a sequence of water body turbidity time series features is used as a sequence of environmental impact key features, which are input into an environmental interference analysis network based on semantic matching to obtain a comprehensive representation of the environmental interference water body spectral semantic response;

[0012] Based on the comprehensive representation of the semantic response of the spectrum of the environmentally disturbed water body, a water quality monitoring result is determined, and the water quality monitoring result is used to represent the water quality category of the monitored water body.

[0013] According to another aspect of the present application, a water quality monitoring system based on an AI optical sensor is provided, comprising:

[0014] A monitored water body spectral data acquisition module, used for acquiring spectral data of the monitored water body through an optical sensor;

[0015] A multi-dimensional data acquisition module for the monitored water body, which monitors the water temperature data and water turbidity data of the monitored water body in real time through a sensor group;

[0016] A water body multi-dimensional time series feature acquisition module, configured to acquire a sequence of water body temperature time series features and a sequence of water body turbidity time series features based on the water temperature data and water body turbidity data of the monitored water body monitored in real time;

[0017] A water body spectral semantic feature extraction module, configured to extract water body spectral semantic features based on the spectral data of the monitored water body;

[0018] A module for comprehensively representing the semantic response of water body spectra to environmental interference is configured to use the semantic features of the water body spectra as reference information matching features, and a feature sequence consisting of a sequence of water body temperature time series features and a sequence of water body turbidity time series features as a sequence of environmental impact key features, and input them into an environmental interference analysis network based on semantic matching to obtain a comprehensive representation of the semantic response of water body spectra to environmental interference;

[0019] The water quality classification module of the monitored water body is used to determine the water quality monitoring results based on the comprehensive representation of the spectral semantic response of the environmental interference water body, and the water quality monitoring results are used to represent the water quality category of the monitored water body.

[0020] Compared with the prior art, a water quality monitoring method and system based on an AI optical sensor provided by the present application monitors water body parameters to track the water body environmental conditions that change over time and cause spectral signal interference. In addition, by establishing a dynamic mapping between spectral features and environmental time series features, spatio-temporal decoupling of interfering components is achieved. That is, by learning the mapping relationship between the time series features of water body parameters and spectral interference, environmental interference analysis is carried out in the form of feature matching, thereby solving the problem that the influence of environmental parameters on the spectrum has a time delay, so as to more accurately predict the spectral interference at the current moment, and eliminate or weaken the interference of water body environmental factors on the spectral signal from the original spectral data, thereby obtaining purer water body component spectral feature information and water quality category information. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other objects, features, and advantages of the present application will become more obvious by describing the embodiments of the present application in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 It is a flowchart of a water quality monitoring method based on an AI optical sensor according to an embodiment of the present application.

[0023] Figure 2 It is a schematic diagram of data flow of a water quality monitoring method based on an AI optical sensor according to an embodiment of the present application.

[0024] Figure 3 It is a flowchart of using the water body spectral semantic feature as the reference information matching feature, and using the feature sequence composed of the sequence of water body temperature time series features and the sequence of water body turbidity time series features as the sequence of environmental impact key features in the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application, and inputting it into the environmental interference analysis network based on semantic matching to obtain the comprehensive characterization of the environmental interference water body spectral semantic response.

[0025] Figure 4 It is a flowchart of calculating the monomer semantic weight of each water body spectral monomer semantic matching score coding vector in the set of water body spectral monomer semantic matching score coding vectors respectively to obtain the set of water body spectral semantic monomer matching semantic self-attention weights in the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application.

[0026] Figure 5 It is a system block diagram of a water quality monitoring system based on an AI optical sensor according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. Like reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0028] As used herein, the term "exemplary" means "serving as an example, embodiment, or illustration". Any embodiment described herein as "exemplary" should not be construed as being superior or better than other embodiments.

[0029] In addition, for a better description of the present application, numerous specific details are given in the following detailed description. Those skilled in the art should understand that the present application can also be implemented without some specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0030] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, "a plurality of" means two or more unless otherwise specifically defined.

[0031] Global water environment problems have promoted the development of water quality monitoring technology towards intelligence and real-time. Although traditional chemical analysis methods have high precision, they are time-consuming and difficult to achieve real-time in-situ monitoring. Water quality monitoring technology based on optical sensors has become a research hotspot due to its advantages such as non-contact, fast response, and continuous monitoring. However, optical sensors are affected by environmental factors such as temperature and turbidity changes, which can cause baseline drift, scattering effects, and changes in absorption intensity of spectral signals, thereby interfering with the spectral information of water components and affecting monitoring accuracy. It should be understood that water temperature changes mainly affect the spectral baseline drift and the intensity of absorption peaks. Turbidity fluctuations mainly cause enhanced light scattering, resulting in a decrease in the overall intensity of the spectrum and a decrease in the signal-to-noise ratio. Traditional methods such as multiple linear correction and baseline deduction are difficult to effectively decouple this complex non-linear interference. Especially under the condition of continuous fluctuations of environmental parameters, existing models generally have the problem of feature drift, seriously affecting monitoring accuracy and the generalization ability of the model.

[0032] In view of the above technical problems, in the technical solution of the present application, a water quality monitoring method based on an AI optical sensor is proposed, which can track the changes of water environment conditions over time by monitoring water body parameters (such as temperature, turbidity), and these changes are the root causes of spectral signal interference. In addition, by establishing a dynamic mapping between spectral features and environmental time series features, the spatio-temporal decoupling of interference components is achieved. That is to say, this solution learns the mapping relationship between the time series features of water body parameters and spectral interference, and analyzes environmental interference in the form of feature matching, thereby solving the problem of time delay in the influence of environmental parameters on the spectrum, so as to more accurately predict the spectral interference at the current moment, and eliminate or weaken the interference of water environment factors on the spectral signal from the original spectral data, so as to obtain purer water body component spectral feature information and water quality category information.

[0033] The present application proposes a water quality monitoring method based on an AI optical sensor. Figure 1 It is a flowchart of the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application includes: S110, collecting spectral map data of the monitored water body through an optical sensor; S120, monitoring the water temperature data and water turbidity data of the monitored water body in real time through a sensor group; S130, based on the water temperature data and water turbidity data of the monitored water body monitored in real time, obtaining a sequence of water temperature time series features and a sequence of water turbidity time series features; S140, extracting water body spectral semantic features based on the spectral map data of the monitored water body; S150, using the water body spectral semantic features as reference information matching features, and using the feature sequence composed of the sequence of water temperature time series features and the sequence of water turbidity time series features as the sequence of environmental impact key features, and inputting them into an environmental interference analysis network based on semantic matching to obtain a comprehensive characterization of environmental interference water body spectral semantic response; S160, determining a water quality monitoring result based on the comprehensive characterization of environmental interference water body spectral semantic response, and the water quality monitoring result is used to represent the water quality category of the monitored water body.

[0034] In the above water quality monitoring method based on an AI optical sensor, in step S110, spectral map data of the monitored water body is collected by the optical sensor. It should be understood that collecting spectral map data of the monitored water body by the optical sensor is a non-contact monitoring method. This method can obtain a large amount of information without disturbing the natural state of the water body. The basic principle of the optical sensor is to utilize the interaction between light of a specific wavelength and the components of the water body, and to reflect the presence and concentration of different substances in the water body by measuring changes in phenomena such as absorption, scattering, or fluorescence. This technology relies on each chemical substance having its unique spectral characteristics, i.e., "fingerprints", under specific conditions, so as to accurately identify and quantify various pollutants. When the optical sensor emits light into the water body, photons will interact with the dissolved substances and suspended particles in the water. These interactions include processes such as absorption, scattering, and reflection. The optical signal finally returned to the sensor contains rich information. For example, certain organic compounds may show absorption peaks in the ultraviolet or visible light range, while inorganic ions may have significant absorption characteristics in the near-infrared region. In addition, by analyzing the intensity and angular distribution of the scattered light, information about the particle size and concentration in the water can be inferred. Thus, collecting water body spectral map data with the help of an optical sensor can not only achieve rapid and continuous monitoring of multiple water quality parameters, but also effectively avoid the problems of long time consumption, high cost, and cumbersome operation existing in traditional sampling and analysis methods. Specifically, a highly sensitive detector is used to capture the optical signal after passing through the water body. Such detectors often have a wide response band and can effectively record the subtle changes caused by the water body. For example, photodiode arrays or charge-coupled devices (CCDs) are often used in such applications because they can obtain a large amount of spectral information in a short time. Once the original optical signal is captured, it will enter a complex signal processing stage. In this stage, the initially obtained data will be converted into digital form and corrected and optimized through a series of algorithms to eliminate the influence of noise interference and other non-target factors. Especially considering the inevitable deviations in the measurement results caused by factors such as temperature fluctuations and turbidity changes in the actual operation environment, appropriate mathematical models must be adopted to compensate for these influences to ensure that the finally obtained spectral map accurately reflects the true condition of the water body. To further improve the measurement accuracy, modern optical sensor systems also integrate an automatic calibration function, allowing the device to self-adjust regularly according to preset standard samples to ensure long-term stability. In addition, with the development of optical fiber technology, it is now possible to introduce and export light through thin and flexible optical fibers to inaccessible waters, greatly expanding the application scenarios of this technology. During the entire collection process, in order to cover as wide a range of water quality parameters as possible, it is usually necessary to repeat measurements multiple times at the same location, each time using a different incident angle or polarization state to collect more-dimensional information.All of the above measures work together to enable water quality monitoring based on optical sensors to not only achieve rapid response but also provide continuous and real-time data streams, which is crucial for timely detection and response to potential water quality problems. In short, with the help of precise optical sensors, researchers can explore the internal structure of water bodies and their dynamic changes at an unprecedented level of detail, providing strong support for environmental protection.

[0035] In the above water quality monitoring method based on an AI optical sensor, in step S120, the water temperature data and water turbidity data of the monitored water body are monitored in real time by a sensor group. It should be understood that temperature, as one of the key parameters affecting physical, chemical, and biological processes in water bodies, its changes can significantly affect the dissolved oxygen concentration, pH value, and the behavior of pollutants in water. By continuously monitoring the water body with a high-precision temperature sensor, the changing trends of these small but important environmental variables can be accurately captured, providing basic data support for water quality analysis. For example, under different seasons or weather conditions, water temperature fluctuations may cause changes in the growth cycle of algae, thereby affecting the balance of the entire ecosystem; in addition, industrial wastewater discharge may also lead to abnormal increases in the temperature of local water areas. This thermal pollution not only directly threatens the living environment of aquatic organisms but may also cause changes in the solubility of harmful substances in water, increasing the risk of environmental pollution. At the same time, water turbidity is an important indicator for measuring water quality transparency, which reflects the quantity and size distribution of suspended particles in water. High turbidity usually means the presence of a large amount of sediment, organic matter, or other impurities. This not only reduces the aesthetic value of the water body but, more importantly, affects the light propagation path in water, reducing the amount of light required for underwater plants to carry out photosynthesis and inhibiting the development of primary productivity in the ecosystem. In addition, an increase in turbidity is often accompanied by an increase in the microbial load because many pathogens can attach to suspended particles and survive and spread. By using an advanced turbidity sensor to monitor water turbidity in real time, not only can water quality deterioration phenomena caused by rainstorms, soil erosion, or human activities be detected in a timely manner, but it also helps to evaluate the effectiveness of water treatment facilities and ensure the safety of drinking water sources. Furthermore, by combining the temperature and turbidity data, more in-depth water quality information can be obtained. This is because there is a complex interaction relationship between the two: for example, in warm seasons, as the air temperature rises, the water temperature on the surface of rivers and lakes increases, accelerating the release rate of nutrients from sediments into the water body, promoting the growth of plankton, and thus indirectly increasing water turbidity; on the contrary, in cold seasons, low temperatures may slow down this process, and ice cover reduces external disturbances, making the water body remain relatively clear. In addition, when a flood event occurs, a large amount of sediment is washed into the river channel, significantly increasing the water turbidity in a short period of time and also changing the local heat exchange pattern, resulting in abnormal fluctuations in water temperature. It can be seen that by synchronously monitoring these two key parameters, the laws of water quality changes can be more accurately simulated and predicted, providing a scientific basis for water resource management decisions. It should be noted that in order to ensure the high accuracy and reliability of the obtained data, modern sensor groups usually integrate a variety of advanced technologies.For example, in terms of temperature measurement, highly sensitive elements made of platinum resistors, thermocouples or semiconductor materials are adopted, which can maintain good linear response characteristics within a wide working range; while in turbidity detection, the principle of scattered light or the method of comparing transmitted light intensities is used, in combination with a precise optical system and signal processing algorithms, which can give reliable readings even in the face of extremely low turbidity levels. At the same time, considering the long-term stability requirements in the field deployment environment, the sensor housing is generally made of corrosion-resistant and impact-resistant materials, and is equipped with a self-cleaning device to prevent measurement errors caused by the formation of biofilms. All these design considerations together ensure that the sensor group can operate stably under various complex conditions, laying a solid foundation for continuously obtaining high-quality water temperature and turbidity data. Finally, with the help of Internet of Things technology, these scattered sensor nodes can upload the collected information to the data center in real time to provide strong technical support.

[0036] In the embodiment of the present application, step S130, based on the water temperature data and water turbidity data of the monitored water body monitored in real time, obtaining a sequence of water temperature time series characteristics and a sequence of water turbidity time series characteristics, includes: S131, after performing data cleaning and denoising processing on the water temperature data and water turbidity data of the monitored water body, dividing them into a time series set of water temperature data and a time series set of water turbidity data according to the water body parameter sample dimension; S132, respectively passing the time series set of water temperature data and the time series set of water turbidity data through a time series feature extractor based on Transformer-TCN to obtain a sequence of water temperature time series feature vectors as the sequence of water temperature time series characteristics and a sequence of water turbidity time series feature vectors as the sequence of water turbidity time series characteristics.

[0037] Specifically, in step S131, after performing data cleaning and denoising on the water temperature data and water turbidity data of the monitored water body, they are divided into a time series set of water temperature data and a time series set of water turbidity data according to the dimension of water body parameter samples. It should be understood that in the actual water quality monitoring scenario, the raw data collected by the optical sensor inevitably contains noise interference. For example, factors such as bubbles generated by water flow, attachments on the sensor surface, or electronic signal fluctuations may introduce outliers or high-frequency noise. If such noise is not processed, it will be amplified through the feature extraction process, further masking the true variation law of water body parameters. Therefore, in the technical solution of this application, after further performing data cleaning and denoising on the water temperature data and water turbidity data of the monitored water body, they are divided into a time series set of water temperature data and a time series set of water turbidity data according to the dimension of water body parameter samples. The data cleaning process can use methods such as sliding window filtering and outlier removal to effectively remove sudden interference signals and retain the low-frequency variation components related to the true dynamics of the water body environment. Subsequently, since the change in water temperature mainly affects the spectral baseline drift and the intensity of absorption peaks, and the turbidity fluctuation mainly causes an increase in light scattering, resulting in a decrease in the overall spectral intensity and a decrease in the signal-to-noise ratio, the cleaned data is further divided into independent water temperature time series sets and turbidity time series sets according to parameter types, providing a structural basis for subsequent capture of different parameter time series features and establishment of parameter-specific interference weights in the semantic matching network.

[0038] Specifically, step S132 passes the time series set of water temperature data and the time series set of water turbidity data through a Transformer-TCN-based time series feature extractor, respectively, to obtain a sequence of water temperature time series feature vectors as the sequence of water temperature time series features, and a sequence of water turbidity time series feature vectors as the sequence of water turbidity time series features. It should be understood that, given that traditional CNN models have difficulty capturing the long-term trends and short-term mutations of parameters such as temperature and turbidity over time due to their fixed-scale local receptive fields, relying solely on the Transformer's global attention mechanism can easily overlook the local temporal patterns of transient scattering interference in turbidity fluctuations. It is worth noting that the impact of water temperature changes on spectral baseline drift often exhibits a slow cumulative effect. For example, diurnal temperature fluctuations caused by summer sunshine can last for hours. In contrast, scattering interference caused by turbidity changes has a rapid response characteristic. For example, a surge in suspended matter concentration caused by rainstorm runoff can significantly change the spectral signal-to-noise ratio within minutes. Therefore, in the technical solution of the present application, the time series set of water temperature data and the time series set of water turbidity data are further processed through a Transformer-TCN-based time series feature extractor to obtain a sequence of water temperature time series feature vectors and a sequence of water turbidity time series feature vectors. Through the processing of the Transformer-TCN-based time series feature extractor, the long-range dependency of water temperature can be modeled using a self-attention mechanism using a dual-channel architecture, and the multi-scale local time series features in turbidity fluctuations can be captured using the dilated causal convolution kernel stack of the TCN module. This complementary design can simultaneously analyze the gradual evolution of temperature influence and the sudden spatiotemporal pattern of turbidity interference.

[0039] In an embodiment of the present application, step S140 extracts water spectral semantic features based on the spectral data of the monitored water body, including: passing the spectral data of the monitored water body through a spectral semantic feature extractor based on a dilated convolutional neural network model to obtain a water spectral semantic feature vector as the water spectral semantic feature. It should be understood that baseline drift and scattering effects caused by environmental factors often manifest as overall morphological distortion of the spectral curve, while absorption peak shifts and intensity changes of water components are reflected in local band characteristics. Due to the limited receptive field of a fixed scale, traditional convolutional neural networks have difficulty simultaneously capturing the multi-scale correlation between wide-range baseline fluctuations and narrowband absorption peaks. Therefore, the spectral data of the monitored water body is further passed through a spectral semantic feature extractor based on a dilated convolutional neural network model to obtain a water spectral semantic feature vector. The spectral semantic feature extractor based on the dilated convolutional neural network model introduces a controllable dilation rate to expand the feature receptive field without increasing the computational complexity, which helps capture and analyze the complex coupling characteristic information between interference components and target features in the spectral signal in the frequency domain.

[0040] Figure 3 In the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application, the spectral semantic features of the water body are used as the benchmark information matching features, and the feature sequence composed of the sequence of the water temperature temporal features and the sequence of the water turbidity temporal features is used as the sequence of the environmental impact key features, and the sequence is input into the environmental interference analysis network based on semantic matching to obtain the flow chart of the comprehensive characterization of the environmental interference water body spectral semantic response. As Figure 3As shown, in the embodiments of the present application, in step S150, the water body spectral semantic features are used as the reference information matching features, and the feature sequence composed of the sequence of water body temperature time series features and the sequence of water body turbidity time series features is used as the sequence of environmental impact key features, and they are input into the environmental interference analysis network based on semantic matching to obtain the comprehensive representation of the environmental interference water body spectral semantic response, including: S151, after enhancing the features of the water body spectral semantic feature vector, it is respectively subjected to monomer semantic matching encoding with each environmental impact key feature vector in the sequence of environmental impact key feature vectors to obtain a set of water body spectral monomer semantic matching score encoding vectors; S152, respectively calculating the monomer semantic weights of each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors to obtain a set of water body spectral semantic monomer matching semantic self-attention weights; S153, aggregating the set of water body spectral monomer semantic matching score encoding vectors based on the set of water body spectral semantic monomer matching semantic self-attention weights to obtain a comprehensive representation vector of the environmental interference water body spectral semantic response as the comprehensive representation of the environmental interference water body spectral semantic response. It should be understood that due to the spatio-temporal heterogeneity and dynamic coupling characteristics of the influence of environmental interference on the spectrum, the traditional method of simply splicing the environmental parameter time series features and spectral features directly ignores the time delay difference and local correlation characteristics of the effects of temperature and turbidity on spectral interference. For example, the baseline drift caused by temperature changes may take some time to fully affect the spectral form, and the light scattering attenuation caused by a sudden increase in turbidity may change the transmittance of multiple bands within a certain period of time. This non-linear spatio-temporal dependence relationship requires the model to have the ability to dynamically match environmental time series segments and spectral features. Therefore, in the technical solution of the present application, further, the water body spectral semantic feature vector is used as the reference information query feature vector, and the feature sequence composed of the sequence of water body temperature time series feature vectors and the sequence of water body turbidity time series feature vectors is used as the sequence of environmental impact key feature vectors, and they are input into the environmental interference analysis network based on semantic matching to obtain a comprehensive representation vector of the environmental interference water body spectral semantic response. Through the processing of the environmental interference analysis network based on semantic matching, the water body spectral semantic features can be used as the query benchmark, which is equivalent to establishing a reference system containing core information such as the position and intensity of the absorption peaks of the target components. The temperature and turbidity time series feature sequences, as key vectors, carry the spatio-temporal evolution patterns of environmental disturbances. Through cross-modal correlation modeling at the semantic level, the spatio-temporal decoupling of environmental interference components is achieved. Specifically, the semantic matching network calculates the dynamic matching scores between each environmental time series feature segment and the current spectral feature through the multi-head cross-attention mechanism, and can establish the dynamic mapping relationship and influence between the environmental time series feature domain spectral features in different local time periods, and use the dynamic matching scores to weight this influence. For example, in this spatio-temporal adaptive weight allocation mechanism, the problem of the lag of environmental parameter influence is effectively solved.

[0041] In an embodiment of the present application, in step S151, after enhancing the features of the water body spectral semantic feature vector, the enhanced vector is respectively subjected to monomer semantic matching encoding with each environmental impact key feature vector in the sequence of environmental impact key feature vectors to obtain a set of water body spectral monomer semantic matching score encoding vectors, including: S1511, performing feature enhancement on the water body spectral semantic feature vector based on deconvolution encoding to obtain a strengthened water body spectral semantic feature vector, and the strengthened water body spectral semantic feature vector has the same feature scale as each environmental impact key feature vector in the sequence of environmental impact key feature vectors; S1512, respectively performing monomer semantic matching encoding on the strengthened water body spectral semantic feature vector and each environmental impact key feature vector in the sequence of environmental impact key feature vectors to obtain the set of water body spectral monomer semantic matching score encoding vectors.

[0042] Specifically, in step S1511, performing feature enhancement on the water body spectral semantic feature vector based on deconvolution encoding to obtain a strengthened water body spectral semantic feature vector, and the strengthened water body spectral semantic feature vector has the same feature scale as each environmental impact key feature vector in the sequence of environmental impact key feature vectors, which is represented by the deconvolution encoding feature enhancement formula as:

[0043] ;

[0044] ;

[0045] where is the water body spectral semantic feature vector, is deconvolution encoding, is the deconvolution weight matrix, is the one-norm of the vector, is the strengthened water body spectral semantic feature vector, is the sequence of environmental impact key feature vectors, are respectively the 1st, 2nd, nd, and An environmental impact key feature vector. It should be understood that due to the inherent differences in the original dimension, semantic granularity, and dynamic characteristics between the spectral semantic features and the environmental impact time series features, direct cross-modal matching will lead to misalignment of the feature space, which in turn causes imbalance in weight allocation and distortion of semantic association. Through a learnable non-linear upsampling mechanism, the deconvolution coding dynamically adjusts the feature channels and spatial resolution, making the enhanced water body spectral semantic feature vector fully match the time series feature dimension of the environmental impact key feature sequence in terms of dimensional topological structure. This not only eliminates the information misalignment caused by scale differences, but more importantly, constructs a unified feature semantic space, providing a mathematical basis for subsequent fine-grained spatio-temporal decoupling based on the attention mechanism. This method abandons the mechanical feature alignment method of traditional linear mapping and instead adopts a data-driven adaptive enhancement strategy, which not only retains the fine spectral fingerprints of the target components in the spectral features, but also enhances the representation ability of environmental interference patterns by expanding the feature expression dimension, thus achieving the essential separation of environmental noise and water quality components at the feature level.

[0046] Specifically, in step S1512, each environmental impact key feature vector in the sequence of the enhanced water body spectral semantic feature vector and the environmental impact key feature vector is respectively subjected to monomer semantic matching coding to obtain a set of water body spectral monomer semantic matching score coding vectors, which is represented by the monomer semantic matching coding formula as:

[0047] ;

[0048] where and are respectively a trainable weight matrix and a trainable bias vector, is a vector concatenation operation, is function, is the A water body spectral monomer semantic matching score encoding vector. It should be understood that since the enhanced water body spectral semantic feature vector represents high-dimensional spectral features such as the absorption peak shape and baseline drift of the target water quality components, and the environmental impact key feature vector sequence carries the temporal evolution pattern of temperature and turbidity fluctuations, there are essential differences between the two in terms of feature dimension, semantic granularity, and dynamic characteristics. The monomer semantic matching encoding dynamically aligns the spectral features with the environmental features of each time segment by introducing a multi-head cross-attention mechanism. Its essence is to establish an implicit association network based on semantic similarity. This matching method can not only capture the dynamic coupling relationship between the global spectral features and the local environmental temporal segments but also reveal the differences in the environmental interference response intensity at different wavelength positions through multi-dimensional scores. For example, the matching score in a specific wavelength region may reflect the influence weight of temperature change on the absorption peak intensity, while the score in another region may reflect the correction degree of turbidity fluctuation on the scattering effect. Compared with the traditional scalar similarity metric, this method provides a more refined feature-level decision basis for subsequent self-attention weight assignment by retaining multi-dimensional semantic association information, thus significantly improving the spatio-temporal resolution of environmental interference decoupling.

[0049] Figure 4 A flowchart for calculating the monomer semantic weights of each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors according to the water quality monitoring method based on an AI optical sensor in an embodiment of the present application to obtain a set of water body spectral semantic monomer matching semantic self-attention weights. As Figure 4 shown, in the embodiment of the present application, step S152, calculating the monomer semantic weights of each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors to obtain a set of water body spectral semantic monomer matching semantic self-attention weights, includes: S1521, determining the monomer semantic matching degree of each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors based on the self-distribution characteristics of the feature set of the set of water body spectral monomer semantic matching score encoding vectors to obtain a set of water body spectral monomer semantic matching degrees; S1522, inputting the set of water body spectral monomer semantic matching degrees into a relational gating proxy module to obtain a set of water body spectral semantic monomer matching semantic self-attention weights.

[0050] In an embodiment of the present application, step S1521, based on the self-distribution characteristics of the feature set of the set of water body spectral monomer semantic matching score encoding vectors, determines the monomer semantic matching degrees of each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors to obtain a set of water body spectral monomer semantic matching degrees, including: S1521-1, performing feature optimization expression based on a mapping specification on each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors to obtain a set of optimized water body spectral monomer semantic matching score encoding vectors; S1521-2, calculating the monomer semantic matching degree of each optimized water body spectral monomer semantic matching score encoding vector based on the distribution characteristics of the set of optimized water body spectral monomer semantic matching score encoding vectors to obtain the set of water body spectral monomer semantic matching degrees.

[0051] Specifically, step S1521, based on the self-distribution characteristics of the feature set of the set of water body spectral monomer semantic matching score encoding vectors, determines the monomer semantic matching degrees of each water body spectral monomer semantic matching score encoding vector in the set of water body spectral monomer semantic matching score encoding vectors to obtain a set of water body spectral monomer semantic matching degrees, which is represented by the water body spectral monomer semantic matching degree determination formula as follows:

[0052] ;

[0053] ;

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] Wherein, is the corresponding water body spectral semantic matching potential vector, is the corresponding projection weight matrix, is the corresponding water body spectral semantic matching interaction encoding vector, is the corresponding coupling constant, for example, calculated in the same way as when performing deconvolution enhancement to maintain symmetry, is the corresponding water body spectral semantic matching covariance matrix, is The corresponding calibrated water body spectrum monomer semantic matching score encoding vector is the first in the set of calibrated water body spectrum monomer semantic matching score encoding vectors. A calibrated water body spectrum monomer semantic matching score encoding vector, is the number of vectors in the set of semantic matching score encoding vectors of water body spectrum monomers after calibration, For the A calibrated water body spectrum monomer semantic matching score encoding vector, is the natural exponential function with base e, for function, for The semantic matching degree of the corresponding water body spectrum. It should be understood that due to the intercorrelations and competition between the individual vectors in the set of semantic matching score encoding vectors, evaluating the semantic matching degree of a single vector in isolation will ignore its relative importance in the overall feature space. By incorporating the self-distribution characteristics of the feature set, the model can identify anomalous vectors that significantly deviate from the group distribution (such as extremely high or low matching results caused by noise) and typical vectors that conform to the inherent patterns of the data. For example, when a fluctuation in an environmental parameter causes a strong absorption peak shift in the spectrum, the corresponding matching score vector will exhibit a significantly different distribution pattern from other vectors along the feature dimension. In this case, the model can determine its higher semantic association value through distribution characteristics analysis. This weight adjustment mechanism based on the collective wisdom essentially transforms the feature matching process into an adaptive feature screening and enhancement process, effectively suppressing the negative impact of noise interference on the matching results while highlighting the dominance of key environmental disturbance features in the spectrum. This improves the robustness of environmental disturbance decoupling while preserving spectral details.

[0059] Specifically, in step S1522, the set of semantic matching degrees of the water body spectrum monomers is input into the relational gating agent module to obtain the set of semantic self-attention weights of the water body spectrum semantic monomer matching, which is expressed as follows using the relational gating agent formula:

[0060] ;

[0061] in, is the mask function, is the preset threshold, for The corresponding water body spectral semantic monomer matching semantic self-attention weights. It should be understood that due to the significant non-linearity and dynamic characteristics of the semantic association patterns implicit in the monomer semantic matching degree set, directly mapping it to the self-attention weights may cause key environmental perturbation information to be submerged or noise to be over-amplified. The relational gating proxy module fine-grained dynamically adjusts the matching degree distribution by simulating the activation-inhibition mechanism of biological neurons. In this way, inhibitory gating is implemented for abnormal matching degrees that significantly deviate from the population distribution (such as false high-correlation signals caused by sensor noise), reducing its interference on the weight calculation. On the other hand, for strong semantic association patterns that conform to the physical laws of environmental perturbations (such as the correspondence between temperature changes and specific absorption peak displacements), enhanced weighting is performed to strengthen its dominant role in weight allocation. This dynamic adjustment based on the gating mechanism essentially constructs an adaptive feature selection network, enabling the finally generated self-attention weights to accurately reflect the deep semantic association between environmental perturbations and spectral features, and effectively suppressing noise interference and false associations, thereby significantly enhancing the model's representation ability for complex water quality environments while ensuring computational efficiency.

[0062] Specifically, in step S153, based on the set of the water body spectral semantic monomer matching semantic self-attention weights, the set of the water body spectral monomer semantic matching score encoding vectors is aggregated to obtain an environmental interference water body spectral semantic response comprehensive representation vector as the environmental interference water body spectral semantic response comprehensive representation, which is expressed by the environmental interference water body spectral semantic response comprehensive representation formula as:

[0063] ;

[0064] where is the environmental interference water body spectral semantic response comprehensive representation vector. It should be understood that due to the spatio-temporal heterogeneity of the impact of environmental interference on spectral signals, the feature matching results of a single time segment are difficult to comprehensively represent complex interference patterns. The self-attention mechanism dynamically generates attention weights adapted to the current input data by calculating the global dependence relationships between each monomer matching score vector, enabling the model to adaptively focus on key environmental perturbation features, so as to achieve the efficient integration of multi-dimensional environmental interference features through the dynamic weight allocation mechanism. For example, when the water body temperature fluctuates periodically, the self-attention mechanism will enhance the correlation weight between the temperature time series features in the corresponding period and the spectral absorption peak displacement, while suppressing irrelevant turbidity fluctuation segments. This data-driven weighted aggregation strategy not only breaks through the representation limitations of traditional linear fusion methods, but also effectively eliminates noise interference and false associations through active selection and filtering at the semantic level, finally forming a refined representation of environmental interference components. Compared with direct concatenation or simple weighting, this method significantly enhances the model's representation ability and anti-interference performance for complex water quality environments through dynamic weight allocation.

[0065] In the above water quality monitoring method based on an AI optical sensor, in step S160, based on the comprehensive characterization of the spectral semantic response of the environmentally disturbed water body, the water quality monitoring result is determined. The water quality monitoring result is used to represent the water quality category of the monitored water body, and includes: passing the comprehensive characterization vector of the spectral semantic response of the environmentally disturbed water body through a water quality analyzer based on a classifier to obtain the water quality monitoring result. It should be understood that although the comprehensive characterization vector of the spectral semantic response of the environmentally disturbed water body has stripped environmental noises such as temperature and turbidity, the water quality information it contains is still scattered in a multi-dimensional feature space - for example, the dissolved oxygen concentration may be associated with the morphology of a specific ultraviolet absorption peak, while heavy metal ions are manifested as a combination of multiple absorption valleys in the visible light band. Therefore, the comprehensive characterization vector of the spectral semantic response of the environmentally disturbed water body is further passed through a water quality analyzer based on a classifier to obtain the water quality monitoring result, and the water quality monitoring result is used to represent the water quality category of the monitored water body.

[0066] In summary, the water quality monitoring method based on an AI optical sensor according to an embodiment of the present application is elucidated. It monitors water body parameters to track the time-varying water body environmental conditions that cause spectral signals to be disturbed. In addition, by establishing a dynamic mapping between spectral features and environmental time series features, the spatio-temporal decoupling of interfering components is achieved. That is, by learning the mapping relationship between the time series features of water body parameters and spectral interference, environmental interference analysis is performed in the way of feature matching, thus solving the problem that the influence of environmental parameters on the spectrum has a time delay, so as to more accurately predict the spectral interference at the current moment, and eliminate or weaken the interference of water body environmental factors on the spectral signal from the original spectral data, thereby obtaining purer water body component spectral feature information and water quality category information.

[0067] Figure 5 FIG. [X] is a system block diagram of a water quality monitoring system based on an AI optical sensor according to an embodiment of the present application. As Figure 5As shown, the water quality monitoring system 100 based on an AI optical sensor according to an embodiment of the present application includes: a monitored water body spectrogram data acquisition module 110 for acquiring spectrogram data of the monitored water body through an optical sensor; a monitored water body multi-dimensional data acquisition module 120 for real-time monitoring of the water temperature data and water turbidity data of the monitored water body through a sensor group; a water body multi-dimensional time series feature acquisition module 130 for acquiring sequences of water temperature time series features and sequences of water turbidity time series features based on the water temperature data and water turbidity data of the monitored water body monitored in real time; a water body spectrogram semantic feature extraction module 140 for extracting water body spectrogram semantic features based on the spectrogram data of the monitored water body; an environmental interference water body spectrogram semantic response comprehensive characterization module 150 for using the water body spectrogram semantic features as reference information matching features, and using the feature sequence composed of the sequences of water temperature time series features and sequences of water turbidity time series features as the sequence of environmental impact key features, and inputting the same into an environmental interference analysis network based on semantic matching to obtain an environmental interference water body spectrogram semantic response comprehensive characterization; a monitored water body water quality category module 160 for determining a water quality monitoring result based on the environmental interference water body spectrogram semantic response comprehensive characterization, and the water quality monitoring result is used to represent the water quality category of the monitored water body.

[0068] Here, those skilled in the art can understand that the specific operations of each step in the above water quality monitoring system based on an AI optical sensor have been introduced in detail in the description of the Figures 1 to 4 water quality monitoring method based on an AI optical sensor above, and therefore, the repeated description thereof will be omitted.

[0069] As described above, the water quality monitoring system 100 based on an AI optical sensor according to an embodiment of the present application can be implemented in various terminal devices. In one example, the water quality monitoring system 100 based on an AI optical sensor can be integrated into a terminal device as a software module and / or a hardware module. For example, the water quality monitoring system 100 based on an AI optical sensor can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the water quality monitoring system 100 based on an AI optical sensor can also be one of many hardware modules of the terminal device.

[0070] Alternatively, in another example, the water quality monitoring system 100 based on an AI optical sensor and the terminal device can also be separate devices, and the water quality monitoring system 100 based on an AI optical sensor can be connected to the terminal device through a wired and / or wireless network, and transmit interaction information according to a predefined data format.

[0071] In summary, the water quality monitoring system based on the AI optical sensor according to the embodiments of the present application is elucidated. By monitoring water body parameters, it tracks the time-varying water body environmental conditions that cause interference to the spectral signal. In addition, by establishing a dynamic mapping between spectral features and environmental time series features, spatio-temporal decoupling of interfering components is achieved. That is, by learning the mapping relationship between the time series features of water body parameters and spectral interference, environmental interference analysis is carried out in the form of feature matching, thus solving the problem of time delay in the influence of environmental parameters on the spectrum, so as to more accurately predict the spectral interference at the current moment, and eliminate or weaken the interference of water body environmental factors on the spectral signal from the original spectral data, thereby obtaining purer spectral feature information of water body components and water quality category information.

[0072] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation.

[0073] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0074] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of a combination of hardware and software functional modules.

[0075] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit of the technical solutions of the present invention.

Claims

1. A water quality monitoring method based on AI optical sensor, characterized in that: include: Collect spectral data of the monitored water body through optical sensors; monitoring the water temperature data and water turbidity data of the monitored water body in real time through a sensor group; Based on the real-time monitored water temperature data and water turbidity data of the monitored water body, a sequence of water body temperature time series characteristics and a sequence of water body turbidity time series characteristics are obtained; Extracting spectral semantic features of the water body based on the spectral data of the monitored water body; The water body spectral semantic features are used as the reference information matching features, and the feature sequence composed of the sequence of water body temperature time series features and the sequence of water body turbidity time series features is used as the sequence of environmental impact key features, which are input into the environmental interference analysis network based on semantic matching to obtain a comprehensive representation of the environmental interference water body spectral semantic response, including: after feature enhancement of the water body spectral semantic feature vector, it is subjected to monomer semantic matching encoding with each environmental impact key feature vector in the sequence of environmental impact key feature vectors to obtain a set of water body spectrum monomer semantic matching score encoding vectors; the monomer semantic weight of each water body spectrum monomer semantic matching score encoding vector in the set of water body spectrum monomer semantic matching score encoding vectors is calculated to obtain a set of water body spectrum semantic monomer matching semantic self-attention weights; based on the set of water body spectrum semantic monomer matching semantic self-attention weights, the set of water body spectrum monomer semantic matching score encoding vectors is aggregated to obtain an environmental interference water body spectral semantic response comprehensive representation vector as the environmental interference water body spectral semantic response comprehensive representation; Determining a water quality monitoring result based on the comprehensive representation of the spectral semantic response of the environmental interference water body, wherein the water quality monitoring result is used to represent the water quality category of the monitored water body; After the water body spectrum semantic feature vector is feature enhanced, it is subjected to monomer semantic matching coding with each environmental impact key feature vector in the sequence of environmental impact key feature vectors to obtain a set of water body spectrum monomer semantic matching score coding vectors, including: The water body spectral semantic feature vector is subjected to feature enhancement based on deconvolution coding to obtain an enhanced water body spectral semantic feature vector. The enhanced water body spectral semantic feature vector has the same characteristic scale as each environmental impact key feature vector in the sequence of environmental impact key feature vectors, and is expressed as follows using the deconvolution coding feature enhancement formula: ； ； in, is the water body spectral semantic feature vector, is deconvolution coding, is the deconvolution weight matrix, is the one-norm of the vector, In order to enhance the semantic feature vector of water spectrum, is a sequence of environmental influence key feature vectors, are the first, second, and third in the sequence of environmental impact key feature vectors. and environmental impact key feature vectors; The enhanced water body spectrum semantic feature vector and each environmental impact key feature vector in the sequence of the environmental impact key feature vector are respectively subjected to monomer semantic matching coding to obtain a set of water body spectrum monomer semantic matching score coding vectors, which is expressed as a monomer semantic matching coding formula: ； in, and are the trainable weight matrix and the trainable bias vector, respectively. For vector concatenation operations, for function, is the first in the set of semantic matching score encoding vectors of water body spectrum monomers A water body spectrum monomer semantic matching score encoding vector; The method of calculating the monomer semantic weight of each water body spectrum monomer semantic matching score encoding vector in the set of water body spectrum monomer semantic matching score encoding vectors to obtain a set of water body spectrum semantic monomer matching semantic self-attention weights includes: Based on the feature set self-distribution characteristics of the set of water body spectrum monomer semantic matching score encoding vectors, the monomer semantic matching degree of each water body spectrum monomer semantic matching score encoding vector in the set of water body spectrum monomer semantic matching score encoding vectors is determined to obtain a set of water body spectrum monomer semantic matching degrees, which is expressed as a water body spectrum monomer semantic matching degree determination formula: ； ； ； ； ； ； in, for The corresponding water body spectral semantic matching potential vector, for The corresponding projection weight matrix, for The corresponding water body spectral semantic matching interaction coding vector, for The corresponding coupling constants are calculated in the same way as for deconvolution enhancement to preserve symmetry, for The corresponding water body spectral semantic matching covariance matrix, for The corresponding calibrated water body spectrum monomer semantic matching score encoding vector is the first in the set of calibrated water body spectrum monomer semantic matching score encoding vectors. A calibrated water body spectrum monomer semantic matching score encoding vector, is the number of vectors in the set of semantic matching score encoding vectors of water body spectrum monomers after calibration, For the A calibrated water body spectrum monomer semantic matching score encoding vector, is the natural exponential function with base e, for function, for The semantic matching degree of the corresponding water body spectrum monomer; The set of semantic matching degrees of the water body spectrum monomers is input into the relational gating agent module to obtain the set of semantic self-attention weights of the water body spectrum semantic monomer matching, which is expressed as follows using the relational gating agent formula: ； in, is the mask function, is the preset threshold, for The corresponding water body spectral semantic monomer matches the semantic self-attention weight; The comprehensive characterization formula of the semantic response of the water body spectrum to environmental interference is expressed as: ； in, A comprehensive representation vector for the spectral semantic response of water bodies subjected to environmental disturbances.

2. The water quality monitoring method based on AI optical sensor according to claim 1 is characterized in that: Based on the real-time monitored water temperature data and water turbidity data of the monitored water body, a sequence of water temperature time series characteristics and a sequence of water turbidity time series characteristics are obtained, including: After data cleaning and denoising, the water temperature data and water turbidity data of the monitored water body are divided into a time series set of water temperature data and a time series set of water turbidity data according to the water parameter sample dimension; The time series set of water temperature data and the time series set of water turbidity data are respectively passed through a Transformer-TCN-based time series feature extractor to obtain a sequence of water temperature time series feature vectors as the sequence of water temperature time series features and a sequence of water turbidity time series feature vectors as the sequence of water turbidity time series features.

3. The water quality monitoring method based on AI optical sensor according to claim 2, characterized in that: Based on the spectral data of the monitored water body, the spectral semantic features of the water body are extracted, including: the spectral data of the monitored water body is passed through a spectral semantic feature extractor based on a void convolutional neural network model to obtain a water body spectral semantic feature vector as the water body spectral semantic feature.

4. The water quality monitoring method based on AI optical sensor according to claim 3 is characterized in that: Determining the monomer semantic matching degree of each water body spectrum monomer semantic matching score encoding vector in the set of water body spectrum monomer semantic matching score encoding vectors based on the feature set self-distribution characteristics of the set of water body spectrum monomer semantic matching score encoding vectors to obtain a set of water body spectrum monomer semantic matching degrees, including: Performing feature optimization expression based on a mapping specification on each water body spectrum monomer semantic matching score encoding vector in the set of water body spectrum monomer semantic matching score encoding vectors to obtain a set of optimized expressed water body spectrum monomer semantic matching score encoding vectors; Based on the distribution characteristics of the set of water body spectrum monomer semantic matching score encoding vectors after the optimized expression, the monomer semantic matching degree of each water body spectrum monomer semantic matching score encoding vector after the optimized expression is calculated to obtain the set of water body spectrum monomer semantic matching degrees.

5. The water quality monitoring method based on AI optical sensor according to claim 4 is characterized in that: Based on the comprehensive representation of the semantic response of the spectrum of the environmental interference water body, the water quality monitoring result is determined, and the water quality monitoring result is used to represent the water quality category of the monitored water body, including: passing the comprehensive representation vector of the semantic response of the spectrum of the environmental interference water body through a water quality analyzer based on a classifier to obtain the water quality monitoring result.

6. A water quality monitoring system based on AI optical sensor, characterized in that: Execute the water quality monitoring method based on AI optical sensor as described in any one of claims 1-5.

Images