Remote sensing dynamic monitoring method, device, equipment and storage medium for sewage outlet into river
Patent Information
- Application Number
- CN202411096493.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-12
AI Technical Summary
Traditional sewage outlet monitoring methods in rivers are difficult to achieve comprehensive and dynamic monitoring of specific areas. A single data source cannot accurately describe complex environmental conditions and lack multi-data source fusion technology.
Using the remote sensing dynamic monitoring method of sewage outlets into the river, the initial water quality sensing monitoring data and airborne remote sensing detection data of the target monitoring task label are obtained, feature mining and pyramid analysis are carried out, combined with knowledge integration, and linkage monitoring and annotation data are generated to achieve the goal. Real-time monitoring and early warning analysis of the area.
It has achieved comprehensive coverage and dynamic monitoring of the sewage outlet areas in the river, improved the accuracy of flow state recognition, provided prediction of water quality change trends, and supported environmental management departments to formulate preventive measures.
Smart Images

Figure CN118643466B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of data analysis technology, and specifically relates to a remote sensing dynamic monitoring method, device, equipment and storage medium for sewage outlets into rivers. Background Art
[0002] In the field of environmental protection, monitoring of sewage outlets into rivers is an important part of protecting water resources and improving environmental quality. However, traditional monitoring methods often rely on fixed water quality monitoring stations, making it difficult to achieve comprehensive and dynamic monitoring of specific areas. At the same time, due to the complex and changeable environmental conditions in the sewage outlet area, a single monitoring data source often cannot provide enough information to accurately describe its conditions. Therefore, a technical solution is needed that can integrate multiple data sources and achieve refined description and dynamic monitoring. Summary of the invention
[0003] The present application provides a remote sensing dynamic monitoring method, device, equipment and storage medium for sewage outlets into rivers, which can solve or partially solve the technical problems involved in the above-mentioned background technology.
[0004] The embodiment of the present application provides a remote sensing dynamic monitoring method for a sewage outlet into a river, which is applied to a remote sensing dynamic monitoring device. The method includes: obtaining initial water quality sensor monitoring data and airborne remote sensing detection data with a target monitoring task label; wherein the target monitoring task label corresponds to a target sewage outlet into a river; performing feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data, respectively, to obtain a water quality sensor monitoring vector and an airborne remote sensing detection vector correspondingly; performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, to obtain a three-dimensional water body spatial inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a number of pyramid depths; performing pyramid flow state recognition on the airborne remote sensing detection vector in turn; wherein the pyramid depth of the pyramid flow state recognition is the same as the pyramid depth of the pyramid three-dimensional water body inversion; between each two consecutive layers of flow state recognition, the flow state recognition result obtained by the previous layer of flow state recognition in the two layers of flow state recognition and the three corresponding pyramid depths are inverted. The invention relates to a method for realizing knowledge integration of a three-dimensional water body space inversion vector; when performing the latter layer of flow state identification in the two-layer flow state identification, flow state identification is performed on the airborne remote sensing detection vector and the knowledge feature set after the knowledge integration; the multimodal sewage discharge monitoring vector obtained after the pyramid flow state identification is processed to obtain the task attention vector associated with the target monitoring task label and the linkage monitoring annotation data of the dynamic remote sensing detection trend vector of the airborne remote sensing detection data; based on the obtained linkage monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data, real-time monitoring and early warning analysis are performed on the target river discharge outlet area; wherein, the task attention vector is used to highlight the data features most relevant to the target monitoring task, the data features are used to locate the key water quality problems of the discharge outlet, and the dynamic remote sensing detection trend vector is used to provide a prediction of the future water quality change trend, and to provide a reference for the early formulation of countermeasures to prevent water quality deterioration.
[0005] The embodiment of the present application provides a remote sensing dynamic monitoring device for a sewage outlet into a river, and the remote sensing dynamic monitoring device for a sewage outlet into a river is used to: obtain initial water quality sensor monitoring data and airborne remote sensing detection data with a target monitoring task label; wherein the target monitoring task label corresponds to a target sewage outlet into a river area; perform feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data respectively, and obtain a water quality sensor monitoring vector and an airborne remote sensing detection vector correspondingly; perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and obtain a three-dimensional water body spatial inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths; perform pyramid flow state recognition on the airborne remote sensing detection vector in sequence; wherein the pyramid depth of the pyramid flow state recognition is the same as the pyramid depth of the pyramid three-dimensional water body inversion; between each two consecutive layers of flow state recognition, the flow state recognition result obtained by the previous layer of flow state recognition in the two layers of flow state recognition and the three corresponding pyramid depths are inverted. The invention relates to a method for realizing knowledge integration of a three-dimensional water body space inversion vector; when performing the latter layer of flow state identification in the two-layer flow state identification, flow state identification is performed on the airborne remote sensing detection vector and the knowledge feature set after the knowledge integration; the multimodal sewage discharge monitoring vector obtained after the pyramid flow state identification is processed to obtain the task attention vector associated with the target monitoring task label and the linkage monitoring annotation data of the dynamic remote sensing detection trend vector of the airborne remote sensing detection data; based on the obtained linkage monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data, real-time monitoring and early warning analysis are performed on the target river discharge outlet area; wherein, the task attention vector is used to highlight the data features most relevant to the target monitoring task, the data features are used to locate the key water quality problems of the discharge outlet, and the dynamic remote sensing detection trend vector is used to provide a prediction of the future water quality change trend, and to provide a reference for the early formulation of countermeasures to prevent water quality deterioration.
[0006] An embodiment of the present application provides a remote sensing dynamic monitoring device, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.
[0007] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.
[0008] In the embodiment of the present application, a sewage outlet linkage monitoring technology solution based on multi-source data fusion and pyramid analysis is proposed. By acquiring initial water quality sensor monitoring data and airborne remote sensing detection data with target monitoring task labels, comprehensive coverage and dynamic monitoring of the target river sewage outlet area are achieved. By performing feature mining and pyramid analysis on these data, the embodiment of the present application can describe the water body condition in a refined manner, improve the accuracy of flow state recognition, and obtain comprehensive and accurate linkage monitoring annotation data. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] Figure 1 A flow chart of a remote sensing dynamic monitoring method for sewage outlets into a river provided in an embodiment of the present application.
[0010] Figure 2 A schematic diagram of the structure of a remote sensing dynamic monitoring device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0012] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, in this application, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.
[0013] Figure 1 A method for dynamic remote sensing monitoring of sewage outlets into a river is shown, which is applied to dynamic remote sensing monitoring equipment. The method includes the following steps 101 to 106.
[0014] Step 101, obtaining initial water quality sensor monitoring data and airborne remote sensing detection data with target monitoring task labels; wherein the target monitoring task labels correspond to the target river discharge outlet area.
[0015] In the embodiment of the present application, the remote sensing dynamic monitoring device first receives a monitoring task with a specific target monitoring task label, which corresponds to the area of the sewage outlet into the river that needs to be focused on. In order to comprehensively and accurately obtain the water quality information of the area, the remote sensing dynamic monitoring device simultaneously obtains two types of data: one is the initial water quality sensor monitoring data from the ground water quality monitoring station, which includes key water quality parameters such as water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, and total phosphorus; the other is the airborne remote sensing detection data obtained through the airborne remote sensing platform, which mainly includes high-resolution optical images, thermal infrared images, and possible lidar data, which are used to provide detailed spatial information of the sewage outlet and its surrounding environment.
[0016] Step 102: Perform feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data respectively, and obtain a water quality sensor monitoring vector and an airborne remote sensing detection vector accordingly.
[0017] In the embodiment of the present application, after acquiring the data, the remote sensing dynamic monitoring equipment conducts in-depth feature mining on these two types of data. For the initial water quality sensor monitoring data, the remote sensing dynamic monitoring equipment uses time series analysis, correlation analysis and other methods to extract key information such as the changing trend, periodic characteristics and abnormal fluctuations of water quality parameters to form a water quality sensor monitoring vector. For airborne remote sensing detection data, the remote sensing dynamic monitoring equipment will use image processing technology, spectral analysis technology and possible machine learning methods to extract spatial features such as surface cover type, water body boundary, sewage outlet location, water flow direction, etc. to form an airborne remote sensing detection vector.
[0018] Step 103: Perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and obtain a three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths.
[0019] In order to understand and analyze water quality conditions more intuitively, remote sensing dynamic monitoring equipment performs pyramid three-dimensional water body inversion on water quality sensor monitoring vectors. Pyramid three-dimensional water body inversion involves expanding the one-dimensional water quality parameter time series to three-dimensional space, that is, adding spatial dimensions (such as longitude, latitude) and depth dimensions on the basis of time dimensions. By performing inversion at different pyramid depths, the three-dimensional spatial distribution of water bodies and their water quality conditions at different scales from macro to micro can be obtained, providing rich background information for subsequent flow state identification.
[0020] Step 104: performing pyramid flow state identification on the airborne remote sensing detection vectors in sequence.
[0021] Among them, the pyramid depth of the pyramid flow state identification is the same as the pyramid depth of the pyramid three-dimensional water body inversion; between each two consecutive layers of flow state identification, the flow state identification result obtained by the previous layer of flow state identification in the two layers of flow state identification and the three-dimensional water body space inversion vector corresponding to the pyramid depth are integrated; when performing the flow state identification of the latter layer of the two layers of flow state identification, flow state identification is performed on the airborne remote sensing detection vector and the knowledge feature set after the knowledge integration.
[0022] Next, the remote sensing dynamic monitoring equipment performs pyramid flow state recognition on the airborne remote sensing detection vector. This process aims to identify the water flow state of the sewage outlet at different pyramid depths, such as flow velocity, flow rate, flow direction, etc. The key is that between each two consecutive layers of flow state recognition, the remote sensing dynamic monitoring equipment performs knowledge integration on the flow state recognition results obtained by the previous layer and the three-dimensional water body spatial inversion vector of the corresponding pyramid depth. Knowledge integration is to generate a knowledge feature set containing richer information by fusing two different types of data features, thereby providing more accurate and comprehensive input for the flow state recognition of the next layer.
[0023] Step 105, processing the multimodal pollution monitoring vector obtained after the pyramid flow state identification, and obtaining linkage monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data.
[0024] Finally, the remote sensing dynamic monitoring equipment processes the multimodal sewage monitoring vector obtained after the pyramid flow state recognition, with the goal of generating linkage monitoring annotation data that associates and matches the task attention vector with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data. The core of this step is to integrate all processed and analyzed data to form a comprehensive monitoring report (linked monitoring annotation data). The linkage monitoring annotation data not only contains real-time monitoring information such as the water quality status and water flow status of the sewage outlet area, but also highlights the data features most relevant to the target monitoring task in the form of task attention vectors. At the same time, the dynamic remote sensing detection trend vector predicts the future trend of water quality changes, providing scientific and comprehensive decision-making support for environmental management departments.
[0025] Step 106, based on the obtained linked monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data, real-time monitoring and early warning analysis of the target river discharge outlet area is performed; wherein, the task attention vector is used to highlight the data features most relevant to the target monitoring task, the data features are used to locate key water quality problems at the discharge outlet, and the dynamic remote sensing detection trend vector is used to provide a prediction of future water quality change trends, which is used to provide a reference for the early formulation of countermeasures to prevent water quality deterioration.
[0026] It can be understood that step 106 is the core link of the sewage outlet monitoring task, and its importance is self-evident. This step is based on the linkage monitoring and annotation data of the task attention vector with the target monitoring task label generated by all the previous steps and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data, and performs real-time monitoring and early warning analysis on the target river sewage outlet area. In this process, it is necessary not only to conduct in-depth analysis and mining of the existing data, but also to convert these data into practical monitoring and early warning information to provide scientific and comprehensive decision-making support for environmental management departments.
[0027] First of all, it should be clear that the role of the task attention vector is to highlight the data features that are most relevant to the target monitoring task. These data features, like indicator lights, can help monitoring personnel quickly locate key water quality problems at the sewage outlet. In the previous steps, in-depth feature mining and analysis of water quality sensor monitoring data and airborne remote sensing detection data have been carried out through methods such as pyramid three-dimensional water body inversion and pyramid flow state recognition. The results of these analyses are integrated into the task attention vector to form the most indicative data feature set for the target monitoring task.
[0028] In actual applications, monitoring personnel can quickly understand the current water quality, water flow status and possible pollution problems in the sewage outlet area by viewing the task attention vector. For example, if the task attention vector shows that a certain water quality parameter is abnormally elevated, the monitoring personnel can immediately focus on this parameter and further analyze its causes and possible impacts. Such prominent display of data features not only improves the work efficiency of monitoring personnel, but also ensures the accuracy and pertinence of monitoring work.
[0029] Secondly, the role of the dynamic remote sensing detection trend vector is to provide a prediction of future water quality change trends. In the previous steps, spatial features such as surface cover type, water body boundary, sewage outlet location, and water flow direction have been extracted through feature mining and analysis of airborne remote sensing detection data. These spatial features, combined with the time series analysis of water quality sensor monitoring data, enable predictions of future water quality change trends. The dynamic remote sensing detection trend vector is a concrete manifestation of this prediction result.
[0030] In practical applications, dynamic remote sensing detection trend vectors can provide important decision-making support for environmental management departments. For example, if the dynamic remote sensing detection trend vector shows that the water quality of a certain area may deteriorate in the future, the environmental management department can formulate and implement corresponding response measures in advance, such as adjusting the sewage discharge strategy and adding water purification facilities. Such prediction and early warning capabilities not only improve the response efficiency of environmental management departments, but also reduce the environmental and economic risks that may be caused by water quality deterioration.
[0031] In addition to the task attention vector and dynamic remote sensing detection trend vector, the linkage monitoring annotation data also contains a wealth of other information, such as detailed spatial information of the sewage outlet area, real-time monitoring data of water quality parameters, etc. This information provides monitoring personnel with comprehensive and accurate information on the water quality and flow status of the sewage outlet area. In practical applications, monitoring personnel can conduct more in-depth and comprehensive analysis and evaluation of the sewage outlet area based on this information.
[0032] For example, monitoring personnel can combine the task attention vector and the dynamic remote sensing detection trend vector to assess the risk of water quality deterioration in the sewage outlet area. If the task attention vector shows that a certain water quality parameter has increased abnormally, and the dynamic remote sensing detection trend vector shows that this parameter may continue to increase in the future, then the monitoring personnel can judge that the risk of water quality deterioration in this area is high and report it to the environmental management department immediately. Such risk assessment and reporting capabilities not only improve the timeliness and accuracy of monitoring work, but also provide a more scientific and comprehensive decision-making basis for environmental management departments.
[0033] In addition, the linked monitoring and labeling data can also provide long-term monitoring and evaluation support for environmental management departments. For example, environmental management departments can use these data to track and monitor the water quality conditions in the sewage outlet area for a long time, and analyze its changing trends and possible causes. Such long-term monitoring and evaluation capabilities not only help environmental management departments to have a deeper understanding of the water quality and water flow conditions in the sewage outlet area, but also provide strong data support for them to formulate more scientific and effective environmental protection policies.
[0034] It can be seen that step 106 is the core link for real-time monitoring and early warning analysis of the target river outlet area based on the generated results of all the previous steps. Through the data feature highlighting of the task attention vector and the prediction of future water quality change trends of the dynamic remote sensing detection trend vector, not only the ability to control the water quality status of the outlet area is improved, but also strong technical support and scientific basis are provided for preventing water quality deterioration and protecting the water environment. At the same time, other information in the linkage monitoring annotation data also provides monitoring personnel with comprehensive and accurate information on water quality and water flow status in the outlet area, providing strong data support for more in-depth and comprehensive analysis and evaluation. Such real-time monitoring and early warning analysis capabilities not only help to improve the response efficiency of environmental management departments, but also reduce the environmental and economic risks that may be caused by water quality deterioration, and provide a strong guarantee for achieving sustainable development and green development.
[0035] It is worth mentioning that in the water quality monitoring task performed by the remote sensing dynamic monitoring equipment, the feature mining of step 102 and the pyramid three-dimensional water body inversion of step 103 constitute one of the core inventions of the embodiment of this application. These two steps not only reflect the in-depth application of remote sensing technology in the field of environmental monitoring, but also demonstrate the innovation and breakthrough of data processing and analysis methods.
[0036] Exemplarily, in step 102, the remote sensing dynamic monitoring equipment performs in-depth feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data. This process is the cornerstone of data analysis and determines the accuracy and effectiveness of subsequent processing.
[0037] For the initial water quality sensor monitoring data, the remote sensing dynamic monitoring equipment uses a variety of mathematical and statistical methods. Time series analysis is used to reveal the trend of water quality parameters over time, which helps to identify long-term changes, seasonal fluctuations or emergencies. Correlation analysis is used to explore the relationship between different water quality parameters, which can reveal the complex biogeochemical processes in the water body. Through these methods, the equipment can extract key water quality features from the original water quality sensor monitoring data to form a water quality sensor monitoring vector.
[0038] For airborne remote sensing detection data, the feature mining process is equally complex and sophisticated. Image processing technology is used to enhance the contrast of images and sharpen edges in order to more clearly identify surface features. Spectral analysis is used to distinguish different types of surface cover, such as water bodies, vegetation, soil, etc. In addition, machine learning algorithms are also introduced to automatically identify key spatial features such as sewage outlets and water body boundaries. The combined use of these technologies enables remote sensing dynamic monitoring equipment to extract rich spatial information from high-resolution remote sensing images and form airborne remote sensing detection vectors.
[0039] Exemplarily, in step 103, the remote sensing dynamic monitoring device performs a pyramid three-dimensional water body inversion on the water quality sensor monitoring vector. This innovative method expands the one-dimensional water quality parameter time series to three-dimensional space, providing a new perspective for water quality monitoring.
[0040] The core idea of pyramid 3D water body inversion is to build water quality models at different spatial scales. By performing inversion at different pyramid depths, the device can generate 3D water body spatial distribution and water quality conditions at different scales from macro to micro. This multi-dimensional representation method not only provides spatial distribution information of water bodies, but also reveals the changes in water quality parameters in the vertical direction.
[0041] In the pyramid three-dimensional water body inversion process, the remote sensing dynamic monitoring equipment first constructs an initial three-dimensional water body model based on the time series data and water body depth information in the water quality sensor monitoring vector. Then, by gradually refining the spatial resolution and increasing the depth level, the equipment generates more refined three-dimensional water body spatial inversion vectors at multiple pyramid depths. These vectors not only contain the spatial distribution information of the water body, but also reflect the changes in water quality parameters at different depths.
[0042] The advantage of pyramid 3D water inversion is that it can provide comprehensive water quality information. Traditional water quality monitoring methods often only focus on the water quality status at a specific point or a specific depth, while pyramid 3D water inversion can reveal the water quality distribution of the entire water body in three-dimensional space. This comprehensive information is of great significance for understanding the overall status of the water body, identifying pollution sources, and predicting water quality trends.
[0043] It can be seen that the combination of feature mining in step 102 and pyramid three-dimensional water body inversion in step 103 has brought significant innovation and improvement to the water quality monitoring task.
[0044] First, the application of feature mining technology enables remote sensing dynamic monitoring equipment to extract key water quality and spatial features from massive amounts of raw data. This not only improves the efficiency of data processing, but also ensures the accuracy of subsequent analysis. Through feature mining, the equipment can more accurately identify key information such as the changing trends of water quality parameters, spatial distribution characteristics, and sewage outlets, providing strong data support for water quality monitoring.
[0045] Secondly, the introduction of the pyramid three-dimensional water body inversion method provides a new perspective for water quality monitoring. By constructing a multi-dimensional water quality model, the equipment can fully reveal the water quality status of the water body in three-dimensional space. This comprehensive information is of great significance for understanding the overall status of the water body, identifying pollution sources, and predicting water quality change trends. The application of the pyramid three-dimensional water body inversion method makes water quality monitoring more refined and comprehensive, providing more scientific and comprehensive decision-making support for environmental management departments.
[0046] Finally, the implementation of the core invention demonstrates the powerful application capabilities of remote sensing technology in the field of environmental monitoring. By combining remote sensing technology with advanced data processing methods, remote sensing dynamic monitoring equipment can achieve efficient and accurate water quality monitoring tasks. This not only improves the efficiency and quality of environmental monitoring, but also provides strong technical support for protecting water resources and improving environmental quality. At the same time, the implementation of this core invention also provides useful reference and reference for the application of remote sensing technology in other environmental monitoring tasks.
[0047] In the water quality monitoring task performed by the remote sensing dynamic monitoring equipment, the pyramid flow state recognition and knowledge integration in step 104 constitutes another core invention point of the embodiment of the present application. This step not only reflects the innovative application of remote sensing technology in flow state recognition, but also demonstrates the advanced integration strategy in data processing and analysis methods.
[0048] Exemplarily, in step 104, the remote sensing dynamic monitoring equipment performs pyramid flow state recognition on the airborne remote sensing detection vector. The core of this process is to identify the flow state at different pyramid depths to obtain multi-scale flow information. The key to pyramid flow state recognition lies in its multi-scale characteristics. Similar to the pyramid three-dimensional water body inversion, pyramid flow state recognition is also performed at different spatial scales. This means that at different pyramid depths, the equipment can identify the flow state at different spatial resolutions. This multi-scale analysis method enables the remote sensing dynamic monitoring equipment to have a more comprehensive understanding of the flow conditions in the sewage outlet area, from macroscopic large-scale flow to microscopic small-scale vortex, which can be accurately identified. In the pyramid flow state recognition process, the remote sensing dynamic monitoring equipment uses advanced image processing technology and machine learning algorithms. By processing the image data in the airborne remote sensing detection vector, the equipment can extract key flow features such as flow velocity, flow rate, and flow direction. At the same time, the application of machine learning algorithms enables the equipment to automatically identify complex flow patterns, such as turbulence, vortices, etc., further improving the accuracy of flow state recognition.
[0049] In detail, in the process of pyramid flow state identification, the remote sensing dynamic monitoring equipment also performs an important knowledge integration step. The core idea of this step is to fuse the flow state identification results obtained by the previous layer of flow state identification and the three-dimensional water body spatial inversion vector of the corresponding pyramid depth between each two consecutive layers of flow state identification to generate a knowledge feature set containing richer information. The realization of knowledge integration relies on advanced data fusion technology. The remote sensing dynamic monitoring equipment generates a knowledge feature set containing more comprehensive and accurate information by fusing these different sources and different types of data. This fusion not only improves the accuracy of flow state identification, but also provides more valuable data support for subsequent monitoring and analysis. Specifically, after the previous layer of flow state identification is completed, the remote sensing dynamic monitoring equipment will fuse the identification results with the three-dimensional water body spatial inversion vector of the corresponding pyramid depth. This fusion is achieved by matching and integrating the key features in the two data sets. For example, the identification results of flow velocity can be associated with the water quality parameters of the corresponding depth to reveal the potential relationship between water quality and flow velocity. Through this fusion, the device is able to generate a knowledge feature set that contains richer and more comprehensive information, providing more accurate and comprehensive input for the subsequent layer of traffic status identification.
[0050] It can be seen that, firstly, the multi-scale characteristics of pyramid flow state recognition enable the remote sensing dynamic monitoring equipment to have a more comprehensive understanding of the flow conditions in the sewage outlet area. By performing flow state recognition at different pyramid depths, the equipment can obtain flow information at different scales from macro to micro. This comprehensive information is of great significance for understanding the flow characteristics of water bodies, identifying potential pollution sources, and predicting flow change trends. Secondly, the application of knowledge integration further improves the accuracy of flow state recognition. By fusing the flow state recognition results of the previous layer and the three-dimensional water body spatial inversion vector corresponding to the pyramid depth, the equipment can generate a knowledge feature set containing richer and more comprehensive information. This fusion not only improves the quality and value of the data, but also provides stronger support for subsequent monitoring and analysis. Finally, the implementation of step 104 demonstrates the powerful application ability of remote sensing technology in flow state recognition. By combining remote sensing technology with advanced data processing methods, remote sensing dynamic monitoring equipment can achieve efficient and accurate flow state recognition tasks. This not only improves the efficiency and quality of environmental monitoring, but also provides strong technical support for protecting water resources and improving environmental quality. At the same time, the implementation of this step also provides a useful reference and reference for the application of remote sensing technology in other environmental monitoring tasks.
[0051] On the basis of the above content, step 105, as another core invention point, undertakes the important task of deeply processing the multimodal pollution monitoring vector obtained after the pyramid flow state recognition, and generating the task attention vector with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data to link the monitoring label data. This step not only reflects the innovation of data processing and analysis methods, but also demonstrates the deep application and advanced integration capabilities of remote sensing technology in the field of environmental monitoring.
[0052] 1. Processing of multi-modal pollution monitoring vectors: deep mining and fusion
[0053] In step 105, the remote sensing dynamic monitoring equipment first processes the multimodal sewage monitoring vectors obtained after the pyramid flow state recognition. These vectors contain rich flow state information, water quality parameters and spatial distribution characteristics, which are the key to understanding the environmental conditions of the sewage outlet area.
[0054] The key to processing multimodal pollution monitoring vectors lies in deep mining and fusion. Remote sensing dynamic monitoring equipment uses advanced machine learning algorithms and deep learning models to conduct in-depth analysis of these vectors. By mining the inherent connections and laws between vectors, the equipment can reveal the complex environmental conditions in the sewage outlet area, such as the diffusion pattern of pollutants and the spatiotemporal changes of water quality parameters.
[0055] At the same time, the fusion of multimodal pollution monitoring vectors is also an important part of the processing. Since these vectors come from different data sources and identification processes, there may be redundant, complementary or contradictory information between them. Remote sensing dynamic monitoring equipment integrates and optimizes the key information in these vectors through intelligent fusion strategies to generate a more comprehensive and accurate description of environmental conditions.
[0056] 2. Generation of Task Attention Vectors: Focusing on Target Monitoring Tasks
[0057] In the process of processing the multimodal pollution monitoring vector, the remote sensing dynamic monitoring equipment also generates a task attention vector associated with the target monitoring task label. The core idea of this vector is to focus on the target monitoring task and extract the most relevant information from the complex environmental conditions.
[0058] The generation of task attention vectors relies on advanced attention mechanisms. This mechanism can simulate the selective attention ability of humans when processing information, that is, quickly filter out the most relevant parts of the current task from a large amount of information. Remote sensing dynamic monitoring equipment uses the attention mechanism to match and associate the key information in the multimodal pollution monitoring vector with the target monitoring task label, and generates a task attention vector containing task-related information.
[0059] The generation of task attention vectors is of great significance for improving the accuracy and efficiency of water quality monitoring tasks. By focusing on the target monitoring task, the device can ignore irrelevant information and interference and more accurately identify the environmental conditions and problems in the sewage outlet area. At the same time, the task attention vector also provides more targeted data support for subsequent monitoring and analysis.
[0060] 3. Generation of dynamic remote sensing detection trend vectors: revealing environmental change trends
[0061] In addition to the task attention vector, the remote sensing dynamic monitoring equipment also generates a dynamic remote sensing detection trend vector of the airborne remote sensing detection data. The core idea of this vector is to reveal the changing trend of the environmental conditions in the sewage outlet area and provide support for prediction and decision-making.
[0062] The generation of dynamic remote sensing detection trend vectors depends on time series analysis and prediction models. Remote sensing dynamic monitoring equipment extracts the changing trends and laws of environmental conditions by analyzing and processing the time series information in airborne remote sensing detection data. At the same time, combined with advanced prediction models, the equipment can also predict and estimate environmental conditions in the future.
[0063] The generation of dynamic remote sensing detection trend vectors is of great significance for understanding the environmental change trend in the sewage outlet area, predicting potential environmental problems, and formulating effective response measures. By revealing the changing trend of environmental conditions, the equipment can provide scientific decision-making basis and support for environmental protection departments and managers.
[0064] 4. Generation and application value of joint monitoring annotation data
[0065] Finally, the remote sensing dynamic monitoring equipment merged and optimized the task attention vector and the dynamic remote sensing detection trend vector to generate linkage monitoring annotation data. These data not only contain relevant information of the target monitoring task, but also reveal the environmental change trends and laws in the sewage outlet area.
[0066] The generation of linked monitoring and annotated data has significant application value for water quality monitoring tasks. First, these data provide environmental protection departments and managers with a comprehensive and accurate description of environmental conditions and trend forecasts, which helps them formulate more effective environmental protection measures and response strategies. Secondly, linked monitoring and annotated data also provide more valuable data support for subsequent monitoring and analysis, which helps to deeply understand and solve environmental problems in the sewage outlet area. Finally, the generation of these data also demonstrates the powerful application capabilities and innovation potential of remote sensing technology in the field of environmental monitoring.
[0067] With this design, the multimodal pollution monitoring vector obtained after deep processing of the pyramid flow state recognition generates linkage monitoring annotation data that associates and matches the task attention vector with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data. The implementation of this step not only demonstrates the innovation and application capabilities of data processing and analysis methods, but also reflects the deep application and advanced integration capabilities of remote sensing technology in the field of environmental monitoring. Through the implementation of this step, remote sensing dynamic monitoring equipment can provide more comprehensive and accurate data support and analysis results for water quality monitoring tasks, and provide strong technical support and decision-making basis for protecting water resources and improving environmental quality.
[0068] In summary, the embodiment of the present application achieves accurate monitoring of the target river discharge outlet area by acquiring the initial water quality sensor monitoring data and airborne remote sensing detection data with target monitoring task labels. By performing feature mining on these data, water quality sensor monitoring vectors and airborne remote sensing detection vectors are obtained, which provide a basis for subsequent in-depth analysis. Further, the embodiment of the present application uses the pyramid three-dimensional water body inversion technology to process the water quality sensor monitoring vector, obtains the three-dimensional water body space inversion vector at several pyramid depths, and realizes a refined description of the water body condition. At the same time, by performing pyramid flow state recognition on the airborne remote sensing detection vector and combining the knowledge integration of the three-dimensional water body space inversion vector, the embodiment of the present application improves the accuracy of flow state recognition and obtains a multi-modal sewage discharge monitoring vector. Finally, by processing these vectors, the embodiment of the present application obtains the linkage monitoring annotation data of the task attention vector and the dynamic remote sensing detection trend vector associated with the target monitoring task label, which provides a comprehensive and accurate description of the environmental status and trend prediction for environmental protection, and helps to formulate more effective environmental protection measures and response strategies.
[0069] In the water quality monitoring task performed by the remote sensing dynamic monitoring equipment, pyramid three-dimensional water body inversion of the water quality sensor monitoring vector is a core step, which aims to extract more detailed and multi-dimensional water body information from the original water quality sensor monitoring data, and provide a solid foundation for subsequent environmental status analysis and prediction. The specific implementation method of this step will be described in detail through two examples below.
[0070] Embodiment 1, the pyramid three-dimensional water body inversion is performed on the water quality sensor monitoring vector, and the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data is obtained at a plurality of pyramid depths, including: performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector through a plurality of feature inversion processing branches in a pre-debugged three-dimensional water body inversion algorithm, and the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data is obtained at a plurality of pyramid depths; wherein, when the first layer of three-dimensional water body inversion is performed on the water quality sensor monitoring vector, the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data is obtained through the plurality of feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm; The first feature inversion processing branch in the feature inversion processing branch performs a first-layer three-dimensional water body inversion on the water quality sensor monitoring vector to obtain a three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at the first layer; when performing an X-th layer of three-dimensional water body inversion on the water quality sensor monitoring vector, the X-th feature inversion processing branch in the plurality of feature inversion processing branches performs an X-th layer of three-dimensional water body space inversion vector on the X-1-th layer to obtain a three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at the X-th layer; X is an integer greater than 1.
[0071] Embodiment 2, the feature inversion processing branch is a first linear knowledge embedding subnet, and the first linear knowledge embedding subnet includes a linear knowledge embedding core. Then, the water quality sensor monitoring vector is subjected to a pyramid three-dimensional water body inversion through several feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm, and the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at several pyramid depths is obtained, including: by calling the linear knowledge embedding core in each first linear knowledge embedding subnet in the several feature inversion processing branches, the water quality sensor monitoring vector is subjected to a pyramid three-dimensional water body inversion, and the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at several pyramid depths is obtained.
[0072] Example 1 describes in detail how to perform pyramid three-dimensional water body inversion on water quality sensor monitoring vectors through several feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm. In this process, the concept of pyramid is introduced, which represents data representations of different levels and different precisions. Each layer of the pyramid corresponds to a three-dimensional water body space inversion vector at a specific depth, and these vectors together constitute a comprehensive description of the water body condition under the target monitoring task label. For example, when the remote sensing dynamic monitoring equipment performs the first layer of three-dimensional water body inversion on the water quality sensor monitoring vector, it will select the first of several feature inversion processing branches to process these data. The feature inversion processing branch has been specially designed and optimized to extract the most basic and intuitive water body features for the original water quality sensor monitoring data, thereby generating the first layer of three-dimensional water body space inversion vectors. This vector is a preliminary but relatively detailed description of the water body condition under the target monitoring task label. Next, when the remote sensing dynamic monitoring equipment needs to perform a higher-level (Xth layer) three-dimensional water body inversion on the water quality sensor monitoring vector, it will select the corresponding Xth feature inversion processing branch. This branch will further extract deeper and more complex water body features based on the three-dimensional water body spatial inversion vector generated in the previous layer (X-1 layer). Through this layer-by-layer progression, the remote sensing dynamic monitoring equipment can gradually build a comprehensive description system of the water body conditions under the target monitoring task label from shallow to deep and from coarse to fine.
[0073] Embodiment 2 is a further refinement and concretization of embodiment 1. Here, the feature inversion processing branch is clearly defined as the first linear knowledge embedding subnet, and the first linear knowledge embedding subnet contains the core linear knowledge embedding kernel. This setting makes the process of pyramid three-dimensional water body inversion clearer and more operational. Specifically, when the remote sensing dynamic monitoring equipment needs to perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, it will call the linear knowledge embedding kernels in several feature inversion processing branches (i.e., the first linear knowledge embedding subnet) in sequence. These linear knowledge embedding kernels are carefully designed and pre-debugged, and they can efficiently extract and convert specific features in water quality sensor monitoring data. Through such processing, the remote sensing dynamic monitoring equipment can gradually generate three-dimensional water body space inversion vectors corresponding to different pyramid depths. It is worth noting that "linear" in the embodiments of the present application does not mean simple or elementary. On the contrary, it represents the efficiency and accuracy of these knowledge embedding kernels in processing data. They can convert complex water quality sensor monitoring data into valuable three-dimensional water body space inversion vectors in a linear, easy to understand and operate manner.
[0074] Through the detailed description of these two examples, we can clearly see how the remote sensing dynamic monitoring equipment converts the original water quality sensor monitoring data into a detailed description of the water body condition under the target monitoring task label through the pyramid three-dimensional water body inversion step. This process not only improves the utilization and value of the data, but also provides a more solid data foundation for subsequent environmental status analysis and prediction.
[0075] Through Example 1 and Example 2, firstly, through the innovative data processing method of pyramid three-dimensional water body inversion, the remote sensing dynamic monitoring equipment can describe the water body conditions under the target monitoring task label more finely and comprehensively. This not only improves the utilization and value of the data, but also provides more accurate and comprehensive environmental status information for environmental protection. Secondly, through the introduction and optimization of the feature inversion processing branch (especially the first linear knowledge embedding subnet and its linear knowledge embedding core), the accuracy and efficiency of data processing are further improved. This enables the remote sensing dynamic monitoring equipment to generate more valuable three-dimensional water body spatial inversion vectors in a shorter time, providing more timely and effective data support for subsequent environmental status analysis and prediction.
[0076] In some other preferred embodiments, the pyramid flow state identification is performed on the airborne remote sensing detection vector in sequence, including: performing pyramid flow state identification on the airborne remote sensing detection vector in sequence through a pre-debugged flow state identification algorithm; wherein the flow state identification algorithm includes a plurality of cascaded second linear knowledge embedding subnets, and a time-series focusing subnet is connected between each two consecutive second linear knowledge embedding subnets; the number of the second linear knowledge embedding subnets is the same as the number of the first linear knowledge embedding subnets in the three-dimensional water body inversion algorithm.
[0077] In the environmental monitoring tasks performed by remote sensing dynamic monitoring equipment, pyramid flow state recognition of airborne remote sensing detection vectors is a crucial step. This step aims to extract key information about water flow from airborne remote sensing detection data and further identify different flow states of water bodies.
[0078] In these preferred embodiments, the remote sensing dynamic monitoring device performs pyramid flow state recognition on the airborne remote sensing detection vector through a pre-debugged flow state recognition algorithm. The algorithm has been carefully designed and optimized to accurately extract the characteristics of water flow from the airborne remote sensing detection data and further identify different flow states of the water body.
[0079] Specifically, the flow state recognition algorithm includes a number of cascaded second linear knowledge embedding subnetworks. These second linear knowledge embedding subnetworks are the same in number as the first linear knowledge embedding subnetworks in the three-dimensional water body inversion algorithm mentioned above, which means that there is a corresponding relationship between them. Each second linear knowledge embedding subnetwork is responsible for processing a specific part of the airborne remote sensing detection data and extracting key features about the water body flow from it.
[0080] It is worth noting that each of the two consecutive second linear knowledge embedding subnetworks is connected to a time-series focusing subnetwork. The role of the time-series focusing subnetwork is very important. It can further integrate and process the features extracted by the first two second linear knowledge embedding subnetworks, thereby highlighting those features that are closely related to the changes in the flow state of the water body. Through such processing, the remote sensing dynamic monitoring equipment can more accurately identify the different flow states of the water body.
[0081] In detail, the remote sensing dynamic monitoring equipment will first input the airborne remote sensing detection vector into the first second linear knowledge embedding subnet. This subnet will process the input data and extract preliminary features about the water flow. Then, these features will be passed to the first time series focusing subnet for further integration and processing.
[0082] Next, the processed features are passed to the second second linear knowledge embedding subnet. This subnet further extracts deeper features about the water flow. These features are then passed to the second time-focused subnet for further integration and processing. This process continues until all second linear knowledge embedding subnets and time-focused subnets have been processed.
[0083] Through such progressive and repeated processing, remote sensing dynamic monitoring equipment can gradually build a comprehensive and detailed description system for the flow state of water bodies. This system includes not only the preliminary characteristics of water flow, but also those deep-level characteristics that are closely related to the changes in water flow state. Ultimately, remote sensing dynamic monitoring equipment will identify different flow states of water bodies based on this description system and provide a solid data foundation for subsequent environmental condition analysis and prediction.
[0084] In this way, through the innovative data processing method of pyramid flow state recognition, remote sensing dynamic monitoring equipment can describe the different flow states of water bodies more finely and comprehensively, which not only improves the utilization rate and value of data, but also provides more accurate and comprehensive water flow information for environmental protection departments and managers. The introduction and optimization of the cascaded second linear knowledge embedding subnet and the time-series focusing subnet further improve the accuracy and efficiency of data processing, so that remote sensing dynamic monitoring equipment can generate more valuable water flow state recognition results in a shorter time, providing more timely and effective data support for subsequent environmental status analysis and prediction. In addition, the organic combination of the three-dimensional water body inversion algorithm and the pyramid flow state recognition algorithm is also realized. By keeping the number of the second linear knowledge embedding subnet the same as the number of the first linear knowledge embedding subnet, the coordination and consistency of the two algorithms in processing data are ensured, which not only improves the efficiency of data processing, but also enhances the accuracy and reliability of the recognition results. Finally, by continuously optimizing and improving data processing and analysis methods, it can be expected that remote sensing technology will make greater contributions to environmental protection in the future. For example, by real-time monitoring of water flow status, possible environmental problems such as water pollution or flooding can be discovered and warned in a timely manner, thus providing strong support for protecting the health and safety of humans and ecosystems.
[0085] In the next step, the pre-debugged flow state recognition algorithm sequentially performs pyramid flow state recognition on the airborne remote sensing detection vector, including: when performing the first-level flow state recognition on the airborne remote sensing detection vector, the first second linear knowledge embedding subnet in the plurality of second linear knowledge embedding subnets is used to perform the first-level flow state recognition on the airborne remote sensing detection vector to obtain the first-level water area flow trend vector of the airborne remote sensing detection data after the first-level flow state recognition; after performing the Y-th level flow state recognition on the airborne remote sensing detection vector, the Y-th level water area flow trend vector after the Y-th level flow state recognition and the Y-th first level in the three-dimensional water body inversion algorithm are used through the Y-th time-series focusing subnet in the flow state recognition algorithm. The three-dimensional water body space inversion vector generated by the linear knowledge embedding subnet is subjected to the knowledge integration to obtain the Yth knowledge feature set; Y is an integer greater than 0; when the Zth level flow state recognition is performed on the airborne remote sensing detection vector, the Zth second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets is used to perform the Zth level flow state recognition on the airborne remote sensing detection vector and the Z-1th knowledge feature set generated by the Z-1th time-series focusing subnet to obtain the Zth level water area flow trend vector after the Zth level flow state recognition of the airborne remote sensing detection data; Z is an integer greater than 1; the water area flow trend vector obtained after the last level flow state recognition of the pyramid flow state recognition is determined as the multimodal pollution discharge monitoring vector obtained after the pyramid flow state recognition.
[0086] In the environmental monitoring tasks performed by remote sensing dynamic monitoring equipment, pyramid flow state recognition of airborne remote sensing detection vectors is a core and complex step. The purpose of this step is to extract key information about water flow from airborne remote sensing detection data and gradually build a comprehensive and detailed description system of water flow state in a layered and progressive manner.
[0087] First, when the remote sensing dynamic monitoring device performs the first-level flow state recognition on the airborne remote sensing detection vector, it selects the first of several second linear knowledge embedding subnets to process the data. The second linear knowledge embedding subnet is specially designed and optimized to efficiently extract and convert specific features in the airborne remote sensing detection data. Through such processing, the remote sensing dynamic monitoring device can generate the first-level water area flow trend vector, which is a preliminary description of the airborne remote sensing detection data after the first-level flow state recognition.
[0088] Next, after the remote sensing dynamic monitoring equipment performs a higher-level (Yth layer) flow state recognition on the airborne remote sensing detection vector, it will use the Yth time-series focused subnet in the flow state recognition algorithm. The role of this subnet is very important. It can integrate the knowledge of the water flow trend vector after the Yth layer of flow state recognition and the three-dimensional water body spatial inversion vector generated by the Yth first linear knowledge embedding subnet in the three-dimensional water body inversion algorithm. Such integrated processing can highlight those features that are closely related to the changes in the water body flow state and provide richer and more comprehensive information for subsequent flow state recognition. Through such processing, the remote sensing dynamic monitoring equipment can generate the Yth knowledge feature set, which contains deep-level features about the water body flow state.
[0089] Then, when the remote sensing dynamic monitoring device needs to identify the flow state of the airborne remote sensing detection vector at a higher level (Zth layer), it will select the Zth of several second linear knowledge embedding subnets to process the data. But this time, it not only processes the airborne remote sensing detection vector, but also takes the Z-1th knowledge feature set generated by the Z-1th time-series focused subnet as input. Through such processing, the remote sensing dynamic monitoring device can generate the Zth layer water area flow trend vector, which is a more detailed description of the airborne remote sensing detection data after the Zth layer flow state identification.
[0090] It is worth noting that such progressive and repeated processing will continue until the last layer of the pyramid flow state identification. After the last layer of flow state identification, the remote sensing dynamic monitoring equipment will obtain a water flow trend vector. This vector is the final result of multi-layer processing and integration of rich information. It is considered to be a multimodal sewage monitoring vector. This vector contains not only preliminary features about water flow, but also deep features that are closely related to changes in water flow state. It provides a solid data foundation for subsequent environmental status analysis and prediction.
[0091] In this way, remote sensing dynamic monitoring equipment can make full use of the information in the airborne remote sensing detection data and gradually build a comprehensive and detailed description system for the water flow state. This system not only improves the utilization rate and value of data, but also provides more accurate and comprehensive water flow information for environmental protection departments and managers. At the same time, through progressive and repeated processing, remote sensing dynamic monitoring equipment can also highlight those features that are closely related to the changes in water flow state, providing richer and more effective data support for subsequent environmental status analysis and prediction.
[0092] In detail, first, through the innovative data processing method of pyramid flow state recognition, the remote sensing dynamic monitoring equipment can describe the different flow states of the water body more finely and comprehensively. This not only improves the utilization and value of the data, but also provides more accurate and comprehensive water flow information for environmental protection departments and managers, which helps them better understand the status of the water body and make scientific decisions. Secondly, through the introduction and optimization of the cascaded second linear knowledge embedding subnet and the time-series focused subnet, the embodiment of the present application further improves the accuracy and efficiency of data processing. This enables the remote sensing dynamic monitoring equipment to generate more valuable water area flow trend vectors and multi-modal sewage monitoring vectors in a shorter time, providing more timely and effective data support for subsequent environmental status analysis and prediction. This not only helps environmental protection departments and managers to discover and solve environmental problems in a timely manner, but also helps scientific research institutions and researchers to have a deeper understanding of the dynamic change process of water bodies. In addition, the embodiment of the present application also realizes the organic combination of a three-dimensional water body inversion algorithm and a pyramid flow state recognition algorithm. By keeping the number of the second linear knowledge embedding subnet the same as the number of the first linear knowledge embedding subnet, and ensuring the coordination and consistency of the two algorithms when processing data, the embodiment of the present application improves the efficiency of data processing and enhances the accuracy and reliability of the recognition results. This provides a more solid data foundation for subsequent environmental monitoring and analysis. Finally, the overall implementation of the embodiment of the present application also helps to promote the in-depth application and innovative development of remote sensing technology in the field of environmental monitoring. By continuously optimizing and improving data processing and analysis methods, it can be expected that remote sensing technology will be able to make greater contributions to environmental protection in the future. For example, by real-time monitoring of the flow state of water bodies and generating multimodal sewage monitoring vectors, possible water pollution incidents or environmental problems such as floods can be discovered and warned in a timely manner, thereby providing strong support for protecting the health and safety of humans and ecosystems.
[0093] In some other implementations, the plurality of second linear knowledge embedding subnets include: at least one second linear knowledge embedding subnet used for performing feature compression on the airborne remote sensing detection vector, and at least one second linear knowledge embedding subnet used for performing feature derivation on the airborne remote sensing detection vector; wherein the number of second linear knowledge embedding subnets used for performing feature compression on the airborne remote sensing detection vector is the same as the number of second linear knowledge embedding subnets used for performing feature derivation on the airborne remote sensing detection vector; and the at least one second linear knowledge embedding subnet used for performing feature derivation on the airborne remote sensing detection vector is connected after the at least one second linear knowledge embedding subnet used for performing feature compression on the airborne remote sensing detection vector.
[0094] In the environmental monitoring tasks performed by remote sensing dynamic monitoring equipment, pyramid flow state recognition of airborne remote sensing detection vectors is a complex and critical process. In order to more effectively extract and process information from airborne remote sensing detection data, the remote sensing dynamic monitoring equipment adopts several second linear knowledge embedding subnets, which are carefully designed to perform specific data processing tasks.
[0095] In these embodiments, the plurality of second linear knowledge embedding subnets are clearly divided into two categories: one is a second linear knowledge embedding subnet used for feature compression of airborne remote sensing detection vectors, and the other is a second linear knowledge embedding subnet used for feature derivation of airborne remote sensing detection vectors. Such a division enables the remote sensing dynamic monitoring device to process airborne remote sensing detection data more specifically, thereby extracting more useful and accurate information.
[0096] Specifically, the main task of the second linear knowledge embedding subnetwork used for feature compression of airborne remote sensing detection vectors is to reduce the dimension of the data while retaining the key information in the original data as much as possible. This step is crucial to improving data processing efficiency and the accuracy of subsequent analysis. Through feature compression, remote sensing dynamic monitoring equipment can remove redundant information in airborne remote sensing detection data, thereby simplifying the data model and reducing computational complexity.
[0097] The second linear knowledge embedding subnetwork, which is used to derive features from airborne remote sensing detection vectors, is responsible for generating new features that are helpful for subsequent analysis. These newly generated features may not be directly reflected in the original data, but they are of great significance in revealing the flow status and changing trends of water bodies. Through feature derivation, remote sensing dynamic monitoring equipment can further mine the potential information in airborne remote sensing detection data, providing richer and more comprehensive data support for subsequent environmental status analysis and prediction.
[0098] It is worth noting that the number of these two types of second linear knowledge embedding subnetworks is the same, and there is at least one. This means that while performing feature compression, the remote sensing dynamic monitoring equipment will also perform the same number of feature derivations. This design ensures the balance and comprehensiveness of data processing, taking into account both data simplification and information richness.
[0099] In addition, these second linear knowledge embedding subnetworks used for feature derivation of airborne remote sensing detection vectors are connected after the second linear knowledge embedding subnetwork used for feature compression of airborne remote sensing detection vectors. Such a connection order is meaningful, which ensures that the data has been effectively compressed before feature derivation, thereby improving the efficiency and accuracy of feature derivation. At the same time, such an order also enables remote sensing dynamic monitoring equipment to gradually build a comprehensive and detailed description system for the flow state of water bodies.
[0100] Specifically, when the remote sensing dynamic monitoring device performs the first-level flow state recognition on the airborne remote sensing detection vector, it will first select a second linear knowledge embedding subnet to process the data for feature compression of the airborne remote sensing detection vector. This subnet will compress the input data, remove redundant information, and retain key features. Then, the processed data will be passed to the first time-series focusing subnet for further integration and processing.
[0101] Next, after the first layer of flow state recognition, the remote sensing dynamic monitoring device will select a second linear knowledge embedding subnet to process the data for feature derivation of the airborne remote sensing detection vector. This subnet will perform feature derivation processing on the data processed in the previous step to generate new features that are helpful for subsequent analysis. These newly generated features will be added to the original feature set, thereby enriching the description of the water flow state.
[0102] Then, the data enriched with the feature set will be passed to the second time-series focusing subnet for further integration and processing. This process will continue until all the second linear knowledge embedding subnets and time-series focusing subnets have been processed. After each layer of flow state recognition, the remote sensing dynamic monitoring equipment will alternately use the second linear knowledge embedding subnet of feature compression and feature derivation to process the data, and gradually build a comprehensive and detailed description system of the water flow state.
[0103] In this way, remote sensing dynamic monitoring equipment can make full use of the information in the airborne remote sensing detection data and gradually extract the key features of the water flow state. At the same time, through the alternating use of feature compression and feature derivation, remote sensing dynamic monitoring equipment can also balance the relationship between data simplification and information richness, providing more accurate and comprehensive data support for subsequent environmental status analysis and prediction.
[0104] In some alternative embodiments, before performing the first layer flow state identification on the airborne remote sensing detection vector, the method also includes: generating a water area flow trend disturbance vector of the flow state identification algorithm through a preconfigured disturbance processing branch; loading the water area flow trend disturbance vector and the airborne remote sensing detection vector as input into the flow state identification algorithm; performing the first layer flow state identification on the airborne remote sensing detection vector through the first second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets to obtain the first layer water area flow trend vector of the airborne remote sensing detection data after the first layer flow state identification, including: performing the first layer flow state identification on the airborne remote sensing detection vector and the water area flow trend disturbance vector through the first second linear knowledge embedding subnet to obtain the first layer water area flow trend vector.
[0105] In order to further improve the accuracy and robustness of flow state identification, the remote sensing dynamic monitoring equipment will perform a series of additional steps before performing the first layer of flow state identification on the airborne remote sensing detection vector.
[0106] First, the remote sensing dynamic monitoring equipment will generate the water flow trend disturbance vector of the flow state recognition algorithm through the pre-configured disturbance processing branch. The purpose of this step is to introduce a random and controllable disturbance to increase the diversity and generalization ability of the flow state recognition algorithm. The water flow trend disturbance vector is a special vector that contains randomly generated disturbance values that conform to a certain distribution law. These disturbance values will be added to the airborne remote sensing detection vector, thereby providing more training samples and change scenarios for the flow state recognition algorithm without affecting the overall characteristics of the original data.
[0107] Next, the remote sensing dynamic monitoring equipment will use the generated water flow trend disturbance vector and the original airborne remote sensing detection vector as input and load them into the flow state recognition algorithm. This step realizes the organic combination of the disturbance vector and the original data, providing richer and more diverse input data for subsequent flow state recognition.
[0108] Then, the remote sensing dynamic monitoring device will perform the first-level flow state recognition on the airborne remote sensing detection vector loaded with the disturbance vector through the first second linear knowledge embedding subnet in several second linear knowledge embedding subnets. This step is the core link of flow state recognition. It uses the powerful feature extraction and conversion capabilities of the second linear knowledge embedding subnet to extract the key features of the water body flow state from the input data and generate the first-level water area flow trend vector.
[0109] Specifically, the first second linear knowledge embedding subnet will jointly process the airborne remote sensing detection vector and the water flow trend disturbance vector. It will gradually extract high-level features from the input data through a series of complex linear transformations and nonlinear activation functions. These features contain both the key information in the original airborne remote sensing detection vector and the diversity and generalization capabilities brought by the disturbance vector. Finally, the first second linear knowledge embedding subnet will output a first-layer water flow trend vector, which is a representation of the result of the input data after the first-layer flow state recognition.
[0110] Through such an implementation, the remote sensing dynamic monitoring equipment can increase the diversity and robustness of the algorithm by introducing the water flow trend disturbance vector before identifying the flow state of the airborne remote sensing detection vector. This not only helps to improve the accuracy of flow state identification, but also enables the algorithm to better adapt to water flow changes in different situations. At the same time, by utilizing the powerful feature extraction capability of the second linear knowledge embedding subnet, the remote sensing dynamic monitoring equipment can extract key features about the water flow state from complex airborne remote sensing detection data, providing more accurate and comprehensive data support for subsequent environmental condition analysis and prediction.
[0111] In this way, by introducing a preconfigured disturbance processing branch and generating a water area flow trend disturbance vector, the remote sensing dynamic monitoring device can increase the diversity and generalization ability of the algorithm before the airborne remote sensing detection vector is used for flow state identification. This not only helps to improve the accuracy of flow state identification, but also enables the algorithm to better adapt to water flow changes in different situations, thereby improving the robustness and practicality of the algorithm. By loading the water area flow trend disturbance vector and the original airborne remote sensing detection vector as input into the flow state identification algorithm, the embodiment of the present application realizes the organic combination of the disturbance vector and the original data. This step provides more abundant and diverse input data for subsequent flow state identification, which helps to further improve the recognition ability and accuracy of the algorithm. By using the first second linear knowledge embedding subnet in several second linear knowledge embedding subnets to perform the first layer of flow state identification on the airborne remote sensing detection vector loaded with the disturbance vector, and generate the first layer of water area flow trend vector, the embodiment of the present application realizes the effective processing and analysis of complex airborne remote sensing detection data. This step not only extracts the key features of the water body flow state, but also provides more accurate and comprehensive data support for subsequent environmental condition analysis and prediction.
[0112] In some other alternative embodiments, the characteristic size of the Y-th layer water flow trend vector is consistent with the characteristic size of the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet; the Y-th time-series focusing subnet in the flow state recognition algorithm performs the knowledge integration on the Y-th layer water flow trend vector after the Y-th layer flow state recognition and the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet in the three-dimensional water body inversion algorithm to obtain the Y-th knowledge feature set, including: performing the knowledge integration on the Y-th layer water flow trend vector after the Y-th layer flow state recognition and the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet in the three-dimensional water body inversion algorithm through the Y-th time-series focusing subnet in the flow state recognition .... The three-dimensional water body space inversion vector generated by the Y-th water area flow trend vector and the Y-th first linear knowledge embedding subnet are subjected to feature fusion at the local attention level to obtain a dynamic monitoring fusion vector; the dynamic monitoring fusion vector is subjected to feature extraction processing to obtain a dynamic monitoring dense vector; the dynamic monitoring dense vector is subjected to quantitative mapping to obtain a confidence relationship network between the Y-th water area flow trend vector and the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet; the dynamic monitoring fusion vector is subjected to knowledge integration processing based on the confidence relationship network to obtain the Y-th knowledge feature set.
[0113] Furthermore, the feature extraction processing is performed on the dynamic monitoring fusion vector to obtain a dynamic monitoring dense vector, including: performing local feature extraction processing and global feature extraction processing on the dynamic monitoring fusion vector respectively to obtain a local monitoring dense vector and a global monitoring dense vector respectively; and determining the local monitoring dense vector and the global monitoring dense vector as the dynamic monitoring dense vector.
[0114] In the environmental monitoring tasks performed by remote sensing dynamic monitoring equipment, the processing of water flow trend vectors and three-dimensional water body spatial inversion vectors is a key step. In order to further improve the accuracy and efficiency of knowledge integration, remote sensing dynamic monitoring equipment will perform a series of specific operations.
[0115] First, the remote sensing dynamic monitoring equipment ensures that the characteristic size of the Y-th layer water flow trend vector is consistent with the characteristic size of the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet. This step is to ensure that the two vectors have the same dimension and feature representation when performing knowledge integration, so that effective feature fusion and refinement can be performed.
[0116] Next, the remote sensing dynamic monitoring equipment integrates the Yth layer of water area flow trend vector after the Yth layer of flow state recognition and the three-dimensional water body space inversion vector generated by the Yth first linear knowledge embedding subnet through the Yth time-series focusing subnet in the flow state recognition algorithm. This step is the core link of knowledge integration, which realizes the feature fusion of the two vectors at the local attention level, thereby generating a dynamic monitoring fusion vector.
[0117] Specifically, the Yth time-series focusing subnetwork will jointly process the Yth layer water flow trend vector and the three-dimensional water body space inversion vector generated by the Yth first linear knowledge embedding subnetwork. It will fuse the features of the two vectors at the local attention level through a series of complex attention mechanisms and feature fusion techniques, thereby generating a dynamic monitoring fusion vector that contains the key information of the two vectors.
[0118] Then, the remote sensing dynamic monitoring equipment performs feature extraction processing on the dynamic monitoring fusion vector to obtain a dynamic monitoring dense vector. This step is to further refine and compress the feature information in the dynamic monitoring fusion vector, thereby generating a more compact and efficient feature representation. Specifically, the remote sensing dynamic monitoring equipment will perform local feature extraction processing and global feature extraction processing on the dynamic monitoring fusion vector. The local feature extraction processing aims to extract the local feature information in the vector and generate a local monitoring dense vector; while the global feature extraction processing aims to extract the global feature information in the vector and generate a global monitoring dense vector. Finally, the remote sensing dynamic monitoring equipment determines the local monitoring dense vector and the global monitoring dense vector as the dynamic monitoring dense vector as the input for subsequent processing.
[0119] Next, the remote sensing dynamic monitoring equipment quantifies and maps the dynamic monitoring dense vector to obtain the confidence relationship network between the Y-th layer water flow trend vector and the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet. This step is to quantify the confidence relationship between the two vectors, thereby generating a confidence relationship network. The confidence relationship network reflects the similarity and correlation between the two vectors at the feature level, providing an important basis for subsequent knowledge integration processing.
[0120] Finally, the remote sensing dynamic monitoring equipment performs knowledge integration processing on the dynamic monitoring fusion vector based on the confidence relationship network to obtain the Yth knowledge feature set. This step is the last step to achieve knowledge integration. The remote sensing dynamic monitoring equipment will further process and process the dynamic monitoring fusion vector based on the information in the confidence relationship network to generate a Yth knowledge feature set containing the key feature information of the two vectors. This knowledge feature set is the result of highly refined and integrated raw data, providing more accurate and comprehensive data support for subsequent environmental condition analysis and prediction.
[0121] By designing in this way, by ensuring that the characteristic size of the Y-th layer water flow trend vector is consistent with the characteristic size of the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet, the embodiment of the present application provides an important premise and foundation for subsequent knowledge integration. This step ensures that the two vectors have the same dimension and feature representation when performing feature fusion and refinement, thereby improving the accuracy and efficiency of knowledge integration. By using the Y-th temporal focus subnet to perform feature fusion of the two vectors at the local attention level, the embodiment of the present application realizes the effective extraction and integration of the key feature information of the two vectors. This step not only improves the accuracy of knowledge integration, but also provides more abundant and comprehensive data support for subsequent environmental status analysis and prediction. By performing local feature extraction and global feature extraction processing on the dynamic monitoring fusion vector, the embodiment of the present application further refines and compresses the feature information in the vector. This step generates a more compact and efficient feature representation, which provides more accurate and reliable input data for subsequent quantitative mapping and knowledge integration processing. By performing knowledge integration processing on the dynamic monitoring fusion vector based on the confidence relationship network, the embodiment of the present application realizes a high degree of refinement and integration of the key feature information of the two vectors. This step generates the Yth knowledge feature set containing the key feature information of the two vectors, providing more accurate and comprehensive data support for subsequent environmental condition analysis and prediction. At the same time, this also further reflects the technical innovation and practical application value of the embodiment of the present application in the field of remote sensing dynamic monitoring.
[0122] Under a possible design idea, the initial water quality sensor monitoring data and the airborne remote sensing detection data are feature mined respectively to obtain a water quality sensor monitoring vector and an airborne remote sensing detection vector correspondingly, including: calling a remote sensing detection feature mining network to perform remote sensing detection feature mining on the airborne remote sensing detection data to obtain the airborne remote sensing detection vector; calling a water quality monitoring feature mining network to perform water quality monitoring feature mining on the initial water quality sensor monitoring data to obtain the water quality sensor monitoring vector.
[0123] Combined with the above content, it can be understood that feature mining of initial water quality sensor monitoring data and airborne remote sensing detection data is a crucial step. The purpose of this step is to extract key feature information from the original data for more accurate and efficient analysis and processing. In one possible design idea, the remote sensing dynamic monitoring device achieves this goal by calling a specific feature mining network.
[0124] First, the remote sensing dynamic monitoring equipment will call the remote sensing detection feature mining network to perform remote sensing detection feature mining on the airborne remote sensing detection data. The remote sensing detection feature mining network is a deep learning model specially designed to process remote sensing detection data. It can analyze the complex patterns in the remote sensing detection data and extract useful feature information from it. In this process, the remote sensing detection feature mining network will perform layer-by-layer convolution, pooling, and nonlinear transformation operations on the airborne remote sensing detection data, thereby gradually refining the high-level features in the data. Finally, the network will output an airborne remote sensing detection vector, which contains the key feature information extracted from the airborne remote sensing detection data.
[0125] At the same time, the remote sensing dynamic monitoring equipment will also call the water quality monitoring feature mining network to mine the water quality monitoring features of the initial water quality sensor monitoring data. The water quality monitoring feature mining network is also a deep learning model, but it is specifically used to process water quality sensor monitoring data. The network can analyze specific components and indicators in the water quality sensor monitoring data and extract feature information reflecting the water quality status from it. Similar to the remote sensing detection feature mining network, the water quality monitoring feature mining network will also process and analyze the initial water quality sensor monitoring data layer by layer, and finally output a water quality sensor monitoring vector. This vector contains the key feature information extracted from the initial water quality sensor monitoring data for subsequent water quality analysis and prediction.
[0126] By calling these two feature mining networks, the remote sensing dynamic monitoring equipment can extract key feature information from the airborne remote sensing detection data and the initial water quality sensor monitoring data, and express this information in the form of vectors. These two vectors (airborne remote sensing detection vector and water quality sensor monitoring vector) are the basis for subsequent analysis and processing. They contain the key information in the original data, but at the same time they have been highly refined and compressed, making the subsequent processing and analysis more efficient and accurate.
[0127] In this way, by calling the remote sensing detection feature mining network and the water quality monitoring feature mining network, the remote sensing dynamic monitoring equipment can extract key feature information from the airborne remote sensing detection data and the initial water quality sensor monitoring data. This step is the basis for subsequent analysis and processing. It improves the availability and value of the data, allowing the remote sensing dynamic monitoring equipment to monitor and analyze environmental conditions more accurately. Through the processing of the feature mining network, the remote sensing dynamic monitoring equipment can convert the raw data into a highly refined and compressed vector form. This vector form of data is easier to store and transmit, and is also more convenient for subsequent processing and analysis. This step improves the efficiency of data processing and analysis, allowing the remote sensing dynamic monitoring equipment to respond to environmental changes more quickly and provide accurate environmental monitoring information in a timely manner.
[0128] In other examples, the three-dimensional water body spatial inversion vector includes a risk warning perspective for characterizing a task requirement of the target monitoring task label; the task requirement includes at least one of the following: pollution monitoring requirements, blockage monitoring requirements, and ecological warning requirements.
[0129] In the environmental monitoring tasks performed by remote sensing dynamic monitoring equipment, the generation of three-dimensional water body space inversion vectors is a core link. This vector not only contains various characteristic information of the water body space, but also, in some examples, specifically contains risk warning views for task requirements that characterize the target monitoring task labels. This design enables remote sensing dynamic monitoring equipment to generate three-dimensional water body space inversion vectors while also providing risk warnings for specific task requirements, thereby improving the pertinence and effectiveness of monitoring.
[0130] Specifically, mission requirements are aspects that remote sensing dynamic monitoring equipment needs to pay special attention to when performing target monitoring tasks. These requirements may come from environmental protection departments, water resources management departments or other relevant agencies to guide remote sensing dynamic monitoring equipment to conduct targeted monitoring. In the example, the mission requirements include at least one of the following: pollution monitoring requirements, blockage monitoring requirements, and ecological early warning requirements.
[0131] Pollution monitoring needs refer to the need to monitor pollutants in water bodies. This includes monitoring various substances that may pollute water bodies, such as chemicals, heavy metals, organic matter, etc. By monitoring the concentration and distribution of these pollutants, remote sensing dynamic monitoring equipment can promptly detect water pollution and provide accurate pollution data to environmental protection departments so that timely treatment measures can be taken.
[0132] The demand for blockage monitoring refers to the demand for monitoring blockage in water bodies. This includes monitoring the blockage of rivers, lakes, reservoirs and other water bodies, as well as early warning of factors that may cause blockage. By monitoring the flow rate, water level and other parameters of the water body, remote sensing dynamic monitoring equipment can promptly detect water blockage and provide accurate blockage data for water resource management departments so that timely dredging measures can be taken to ensure the normal flow of water bodies.
[0133] Ecological early warning demand refers to the need for early warning of the ecological environment of water bodies. This includes monitoring the health of the water ecosystem and early warning of factors that may cause damage to the ecosystem. By monitoring the biodiversity, water quality and other parameters of the water body, remote sensing dynamic monitoring equipment can promptly discover problems in the water ecosystem and provide accurate ecological data for environmental protection departments so that timely protection measures can be taken to maintain the ecological balance of the water body.
[0134] In the process of generating the three-dimensional water body space inversion vector, the remote sensing dynamic monitoring equipment will incorporate the risk warning viewpoints of these task requirements. Specifically, the equipment will weight the various characteristic information of the water body space according to the different task requirements, so that the characteristic information related to the task requirements will receive more attention in the vector. At the same time, the equipment will also set different risk warning thresholds according to different task requirements, so that when abnormal conditions related to task requirements are monitored, risk warning signals can be issued in time.
[0135] In this way, the remote sensing dynamic monitoring equipment can not only generate a three-dimensional water body spatial inversion vector containing rich water body spatial feature information, but also carry out targeted risk warning according to specific mission requirements. This design improves the monitoring effect and practicality of remote sensing dynamic monitoring equipment, enabling it to better serve the fields of environmental protection and water resources management.
[0136] In some optional embodiments, the remote sensing dynamic monitoring method of the sewage outlet into the river is implemented by a pre-debugged deep learning network; the deep learning network includes a first deep learning branch and a second deep learning branch; the first deep learning branch includes at least a remote sensing detection feature mining network and a flow state recognition algorithm, and the second deep learning branch includes a water quality monitoring feature mining network and a three-dimensional water body inversion algorithm; the deep learning network is debugged through the following steps: loading the water quality sensor monitoring data samples in the training sample set into the second deep learning branch, and performing water quality monitoring feature mining on the water quality sensor monitoring data samples through the water quality monitoring feature mining network in the second deep learning branch to obtain water quality sensor monitoring vector samples; performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector samples through the three-dimensional water body inversion algorithm in the second deep learning branch to obtain the corresponding water quality sensor monitoring The monitoring task label samples in the data samples are three-dimensional water body space inversion vector samples at several pyramid depths; the airborne remote sensing detection data samples in the training sample set are loaded into the first deep learning branch, and the airborne remote sensing detection data samples are subjected to remote sensing detection feature mining through the remote sensing detection feature mining network in the first deep learning branch to obtain airborne remote sensing detection vector samples; the airborne remote sensing detection vector samples are subjected to pyramid flow state recognition in turn through the flow state recognition algorithm in the first deep learning branch to obtain multimodal pollution monitoring vector samples; the training error of the deep learning network is determined based on the three-dimensional water body space inversion vector samples and the multimodal pollution monitoring vector samples to obtain the network training error of the deep learning network; the network weight variables in the deep learning network are improved based on the network training error to obtain a debugged deep learning network.
[0137] Based on this embodiment, the remote sensing dynamic monitoring method of the sewage outlet into the river performed by the remote sensing dynamic monitoring device is implemented through a pre-debugged deep learning network. The deep learning network is carefully designed and includes two main learning branches: a first deep learning branch and a second deep learning branch. Each branch undertakes a specific task and integrates advanced algorithms and technologies to ensure the accuracy and efficiency of monitoring.
[0138] The first deep learning branch includes at least a remote sensing detection feature mining network and a flow state recognition algorithm. The remote sensing detection feature mining network is a deep learning model specifically used to process and analyze airborne remote sensing detection data. It can deeply mine the hidden features in the data and provide strong support for subsequent flow state recognition. The flow state recognition algorithm is responsible for accurately identifying and monitoring the flow state of the sewage outlet based on the feature information extracted by the remote sensing detection feature mining network.
[0139] The second deep learning branch includes a water quality monitoring feature mining network and a three-dimensional water body inversion algorithm. The water quality monitoring feature mining network is also a deep learning model, but it is specifically used to process and analyze water quality sensor monitoring data. The network can extract key feature information reflecting the water quality status and provide basic data for subsequent three-dimensional water body inversion. The three-dimensional water body inversion algorithm is responsible for performing a three-dimensional inversion of the water body based on the feature information extracted by the water quality monitoring feature mining network, so as to intuitively display the spatial distribution and status of the water body.
[0140] The debugging process of the deep learning network is a complex but crucial step. First, the remote sensing dynamic monitoring device will load the water quality sensor monitoring data samples in the training sample set into the second deep learning branch. Then, the water quality monitoring feature mining network performs feature mining on these data samples to obtain water quality sensor monitoring vector samples. These vector samples contain key feature information in the original data, providing a basis for subsequent three-dimensional water body inversion.
[0141] Next, the remote sensing dynamic monitoring equipment will use the three-dimensional water body inversion algorithm to perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector samples. This process will obtain three-dimensional water body space inversion vector samples at several pyramid depths. These vector samples intuitively show the distribution and status of the monitoring task label samples in the water quality sensor monitoring data samples in the water body space.
[0142] At the same time, the remote sensing dynamic monitoring equipment will also load the airborne remote sensing detection data samples in the training sample set into the first deep learning branch. The remote sensing detection feature mining network performs feature mining on these data samples to obtain airborne remote sensing detection vector samples. These vector samples also contain key feature information in the original data, providing a basis for subsequent traffic state recognition.
[0143] Subsequently, the remote sensing dynamic monitoring equipment will use the flow state recognition algorithm to perform pyramid flow state recognition on the airborne remote sensing detection vector samples. This process will obtain multi-modal sewage discharge monitoring vector samples, which accurately reflect the flow state information of the sewage outlet.
[0144] Finally, the remote sensing dynamic monitoring equipment will determine the training error of the deep learning network based on the three-dimensional water body space inversion vector samples and the multimodal sewage monitoring vector samples. The network training error can be obtained by calculating the difference between the network output and the true label. Then, the remote sensing dynamic monitoring equipment will improve and optimize the network weight variables in the deep learning network according to the network training error to obtain a debugged deep learning network. The debugged network has higher accuracy and stability, and can better serve the remote sensing dynamic monitoring task of the sewage outlet into the river.
[0145] With such a design, remote sensing dynamic monitoring of sewage outlets into rivers is realized through a pre-debugged deep learning network, which improves the accuracy and efficiency of monitoring. The deep learning network can deeply explore and utilize the characteristic information in remote sensing detection data and water quality sensor monitoring data, providing strong support for subsequent monitoring and analysis. Through the application of advanced algorithms such as pyramid three-dimensional water body inversion and pyramid flow state recognition, the embodiment of the present application realizes accurate monitoring and visual display of water body space and sewage outlet flow state. This enables relevant departments and personnel to understand the water body conditions and sewage outlet conditions more intuitively, and provides a strong basis for formulating targeted governance measures. Through the improvement and optimization of training error determination and network weight variables, the embodiment of the present application continuously improves the performance and stability of the deep learning network. This enables remote sensing dynamic monitoring equipment to maintain accurate monitoring effects under changing environmental conditions, and provide continuous and reliable monitoring data support for fields such as environmental protection and water resources management.
[0146] In some independent embodiments, the training error of the deep learning network is determined based on the three-dimensional water body spatial inversion vector samples and the multimodal pollution monitoring vector samples to obtain the network training error of the deep learning network, including: determining the initial training error of the deep learning network based on the multimodal pollution monitoring vector samples and the water quality sensor monitoring data samples to obtain the initial network training error; determining the water body inversion training error of the deep learning network based on the three-dimensional water body spatial inversion vector samples and the multimodal pollution monitoring vector samples to obtain the water body inversion training error; and determining the network training error of the deep learning network by combining the initial network training error and the water body inversion training error.
[0147] In other independent embodiments, the initial training error of the deep learning network is determined based on the multimodal pollution discharge monitoring vector samples and the water quality sensor monitoring data samples to obtain the initial network training error, including: processing the multimodal pollution discharge monitoring vector samples to obtain linked monitoring annotation data samples that are associated and matched with the task attention vector of the monitoring task label sample and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data sample; and determining the offset weight between the linked monitoring annotation data sample and the water quality sensor monitoring data sample as the initial network training error.
[0148] In some other independent embodiments, the water body inversion training error of the deep learning network is determined based on the three-dimensional water body spatial inversion vector samples and the multimodal pollution monitoring vector samples to obtain the water body inversion training error, including: processing the three-dimensional water body spatial inversion vector samples to obtain linkage monitoring training data; processing the multimodal pollution monitoring vector samples to obtain linkage monitoring annotation data samples that are associated and matched with the task attention vector of the monitoring task label sample and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data sample; performing verification analysis on the linkage monitoring training data and the linkage monitoring annotation data samples respectively, and correspondingly obtaining a first verification analysis feature and a second verification analysis feature; determining the quantitative feature difference between the first verification analysis feature and the second verification analysis feature as the water body inversion training error.
[0149] In detail, when the remote sensing dynamic monitoring equipment executes the remote sensing dynamic monitoring method of the sewage outlet into the river, it will determine the training error of the deep learning network based on the three-dimensional water body space inversion vector samples and the multimodal sewage monitoring vector samples, thereby obtaining the network training error of the deep learning network.
[0150] First, the remote sensing dynamic monitoring equipment will determine the initial training error of the deep learning network based on the multimodal pollution monitoring vector samples and the water quality sensor monitoring data samples to obtain the initial network training error. The purpose of this step is to evaluate the performance of the deep learning network on the pollution monitoring task in the initial training stage. Specifically, the remote sensing dynamic monitoring equipment will process the multimodal pollution monitoring vector samples, and through a series of calculations and analyses, obtain the linkage monitoring annotation data samples that are associated with the task attention vector of the monitoring task label sample and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data sample. This processing process aims to extract the key information in the multimodal pollution monitoring vector samples, and associate and match it with the monitoring task label samples and the airborne remote sensing detection data samples for subsequent training and verification analysis. Then, the remote sensing dynamic monitoring equipment will determine the offset weight between the linkage monitoring annotation data samples and the water quality sensor monitoring data samples as the initial network training error. This step evaluates the accuracy of the deep learning network for the pollution monitoring task in the initial training stage by comparing the differences between the linked monitoring annotated data samples and the water quality sensor monitoring data samples.
[0151] Secondly, the remote sensing dynamic monitoring equipment will determine the water body inversion training error of the deep learning network based on the three-dimensional water body space inversion vector samples and the multimodal pollution monitoring vector samples to obtain the water body inversion training error. The purpose of this step is to evaluate the performance of the deep learning network in the water body inversion task. Specifically, the remote sensing dynamic monitoring equipment will process the three-dimensional water body space inversion vector samples, and obtain the linkage monitoring training data through a series of calculations and analyses. This processing process aims to extract the key information in the three-dimensional water body space inversion vector samples for subsequent training and verification analysis. At the same time, the remote sensing dynamic monitoring equipment will also process the multimodal pollution monitoring vector samples to obtain the linkage monitoring annotation data samples of the task attention vector associated with the monitoring task label sample and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data sample. Then, the remote sensing dynamic monitoring equipment will perform verification analysis on the linkage monitoring training data and the linkage monitoring annotation data samples respectively to obtain the first verification analysis feature and the second verification analysis feature. This step evaluates the accuracy of the deep learning network in the water body inversion task by comparing the differences between the linked monitoring training data and the linked monitoring annotated data samples. Finally, the remote sensing dynamic monitoring equipment will determine the quantitative feature difference between the first verification analysis feature and the second verification analysis feature as the water body inversion training error.
[0152] Finally, the remote sensing dynamic monitoring equipment will combine the initial network training error with the water body inversion training error to determine the network training error of the deep learning network. The purpose of this step is to comprehensively evaluate the overall performance of the deep learning network in sewage monitoring tasks and water body inversion tasks. By combining the initial network training error and the water body inversion training error, the remote sensing dynamic monitoring equipment can obtain a comprehensive network training error evaluation result, thereby guiding subsequent network optimization and debugging work.
[0153] In addition, the specific process of determining the initial training error of the deep learning network based on the multimodal pollution monitoring vector samples and the water quality sensor monitoring data samples is as follows: the remote sensing dynamic monitoring equipment will first process the multimodal pollution monitoring vector samples, extract the key information through a series of complex calculation and analysis steps, and associate and match them with the monitoring task label samples and the airborne remote sensing detection data samples. This processing process aims to obtain the linked monitoring labeled data samples that are associated with the task attention vectors of the monitoring task label samples and the dynamic remote sensing detection trend vectors of the airborne remote sensing detection data samples. Then, the remote sensing dynamic monitoring equipment will calculate the offset weight between the linked monitoring labeled data samples and the water quality sensor monitoring data samples, which reflects the degree of difference between the two. Finally, the remote sensing dynamic monitoring equipment will determine this offset weight as the initial network training error as an important indicator for evaluating the accuracy of the deep learning network in the initial training stage for the pollution monitoring task.
[0154] Furthermore, the specific process of determining the water body inversion training error of the deep learning network based on the three-dimensional water body space inversion vector sample and the multimodal pollution monitoring vector sample is as follows: the remote sensing dynamic monitoring equipment will first process the three-dimensional water body space inversion vector sample, extract the key information through a series of calculation and analysis steps, and obtain the linkage monitoring training data. At the same time, the remote sensing dynamic monitoring equipment will also process the multimodal pollution monitoring vector sample to obtain the linkage monitoring annotation data sample of the task attention vector associated with the monitoring task label sample and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data sample. Then, the remote sensing dynamic monitoring equipment will verify and analyze the linkage monitoring training data and the linkage monitoring annotation data samples respectively, and obtain the first verification analysis feature and the second verification analysis feature. These two features respectively reflect the prediction results and the actual situation of the deep learning network in the water body inversion task. Finally, the remote sensing dynamic monitoring equipment will calculate the quantitative feature difference between the first verification analysis feature and the second verification analysis feature, and determine this difference as the water body inversion training error. This error indicator reflects the accuracy of the deep learning network in the water body inversion task and is an important basis for guiding subsequent network optimization and debugging work.
[0155] In this way, through the designed deep learning network and a series of complex calculation and analysis steps, the remote sensing dynamic monitoring of the sewage outlet into the river is realized. The embodiment of the present application not only improves the accuracy and efficiency of monitoring, but also further improves the comprehensiveness and reliability of monitoring by introducing the strategy of multi-task learning and multi-modal data fusion. Through the pre-debugging and training error determination steps of the deep learning network, the embodiment of the present application can continuously optimize and improve the network model to improve its accuracy and stability for sewage monitoring tasks and water body inversion tasks. This enables the remote sensing dynamic monitoring equipment to maintain efficient monitoring performance under complex and changeable environmental conditions. By introducing the strategy of multi-task learning, by considering the sewage monitoring task and the water body inversion task at the same time, comprehensive monitoring of the sewage outlet into the river is achieved. This strategy not only improves the comprehensiveness of monitoring, but also can further improve the accuracy of monitoring through the interaction and constraints between tasks. Through the strategy of multi-modal data fusion, full use is made of data information from two different sources, airborne remote sensing detection data and water quality sensor monitoring data. By integrating and analyzing them, the embodiment of the present application can more accurately reflect the actual situation of the sewage outlet into the river and provide more reliable and comprehensive monitoring data support for relevant departments and personnel.
[0156] Further, Figure 2 Schematic diagram of a remote sensing dynamic monitoring device 200 provided in an embodiment of the present application. Figure 2 The remote sensing dynamic monitoring device 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.
[0157] Alternatively, if Figure 2 As shown, the remote sensing dynamic monitoring device 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.
[0158] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .
[0159] Alternatively, if Figure 2 As shown, the remote sensing dynamic monitoring device 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.
[0160] Optionally, the remote sensing dynamic monitoring device 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application. For the sake of brevity, they are not repeated here.
[0161] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities.
[0162] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.
[0163] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.
[0164] On the basis of the above, based on the same inventive concept, a remote sensing dynamic monitoring device for a sewage outlet into a river is provided, and the remote sensing dynamic monitoring device for a sewage outlet into a river is used to: obtain initial water quality sensor monitoring data and airborne remote sensing detection data with target monitoring task labels; wherein the target monitoring task labels correspond to the target sewage outlet into a river area; perform feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data respectively, and obtain water quality sensor monitoring vectors and airborne remote sensing detection vectors accordingly; perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vectors, and obtain three-dimensional water body spatial inversion vectors of the target monitoring task labels in the initial water quality sensor monitoring data at several pyramid depths; perform pyramid flow state recognition on the airborne remote sensing detection vectors in turn; Among them, the pyramid depth of the pyramid flow state identification is the same as the pyramid depth of the pyramid three-dimensional water body inversion; between each two consecutive layers of flow state identification, the flow state identification result obtained by the previous layer of flow state identification in the two layers of flow state identification and the three-dimensional water body space inversion vector corresponding to the pyramid depth are integrated; when performing the latter layer of flow state identification in the two layers of flow state identification, flow state identification is performed on the airborne remote sensing detection vector and the knowledge feature set after the knowledge integration; the multimodal sewage discharge monitoring vector obtained after the pyramid flow state identification is processed to obtain the linkage monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data.
[0165] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.
[0166] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0167] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the present application, all of which are within the protection of the present application.
Claims
1. A remote sensing dynamic monitoring method for sewage outlets into rivers, characterized in that: Applied to remote sensing dynamic monitoring equipment, the method comprises: Acquire initial water quality sensor monitoring data and airborne remote sensing detection data with target monitoring task labels; wherein the target monitoring task labels correspond to the target river discharge outlet area; Performing feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data respectively, and obtaining a water quality sensor monitoring vector and an airborne remote sensing detection vector correspondingly; Performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and obtaining corresponding three-dimensional water body space inversion vectors of the target monitoring task label in the initial water quality sensor monitoring data at several pyramid depths; The airborne remote sensing detection vector is sequentially subjected to pyramid flow state recognition; wherein the pyramid depth of the pyramid flow state recognition is the same as the pyramid depth of the pyramid three-dimensional water body inversion; between each two consecutive layers of flow state recognition, the flow state recognition result obtained by the previous layer of flow state recognition in the two layers of flow state recognition and the three-dimensional water body space inversion vector corresponding to the pyramid depth are subjected to knowledge integration; when the latter layer of flow state recognition in the two layers of flow state recognition is subjected to flow state recognition on the airborne remote sensing detection vector and the knowledge feature set after the knowledge integration; The multimodal sewage monitoring vector obtained after the pyramid flow state recognition is processed to obtain linkage monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data; Based on the obtained task attention vector associated with the target monitoring task label and the linkage monitoring annotation data of the dynamic remote sensing detection trend vector of the airborne remote sensing detection data, real-time monitoring and early warning analysis are performed on the target river discharge outlet area; wherein the task attention vector is used to highlight the data features most relevant to the target monitoring task, the data features are used to locate key water quality problems at the discharge outlet, and the dynamic remote sensing detection trend vector is used to provide a prediction of future water quality change trends, which is used to provide a reference for the early formulation of countermeasures to prevent water quality deterioration; The pyramid three-dimensional water body inversion is performed on the water quality sensor monitoring vector to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths, including: performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector through a plurality of feature inversion processing branches in a pre-debugged three-dimensional water body inversion algorithm to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths; wherein: when the first layer of three-dimensional water body inversion is performed on the water quality sensor monitoring vector, the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data is obtained through the plurality of feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm; The first feature inversion processing branch in the feature inversion processing branch performs the first layer of three-dimensional water body inversion on the water quality sensor monitoring vector to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at the first layer; when the water quality sensor monitoring vector is subjected to the Xth layer of three-dimensional water body inversion, the Xth feature inversion processing branch in the plurality of feature inversion processing branches performs the Xth layer of three-dimensional water body inversion on the three-dimensional water body space inversion vector of the X-1th layer to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at the Xth layer; X is an integer greater than 1; The feature inversion processing branch is a first linear knowledge embedding subnet, and the first linear knowledge embedding subnet includes a linear knowledge embedding core; the plurality of feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and correspondingly obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths, including: by calling the linear knowledge embedding core in each first linear knowledge embedding subnet in the plurality of feature inversion processing branches, perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and correspondingly obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths; The method sequentially performing pyramid flow state identification on the airborne remote sensing detection vector comprises: sequentially performing pyramid flow state identification on the airborne remote sensing detection vector through a pre-debugged flow state identification algorithm; wherein the flow state identification algorithm comprises a plurality of cascaded second linear knowledge embedding subnets, and a time-series focusing subnet is connected between each two consecutive second linear knowledge embedding subnets; the number of the second linear knowledge embedding subnets is the same as the number of the first linear knowledge embedding subnets in the three-dimensional water body inversion algorithm; The pre-debugged flow state recognition algorithm sequentially performs pyramid flow state recognition on the airborne remote sensing detection vector, including: When the airborne remote sensing detection vector is subjected to the first-layer flow state identification, the airborne remote sensing detection vector is subjected to the first-layer flow state identification through the first second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets, so as to obtain the first-layer water area flow trend vector of the airborne remote sensing detection data after the first-layer flow state identification; After performing the Yth layer flow state recognition on the airborne remote sensing detection vector, the Yth layer water area flow trend vector after the Yth layer flow state recognition and the three-dimensional water body space inversion vector generated by the Yth first linear knowledge embedding subnet in the three-dimensional water body inversion algorithm are subjected to the knowledge integration through the Yth time-series focusing subnet in the flow state recognition algorithm to obtain the Yth knowledge feature set; Y is an integer greater than 0; When the Z-th level flow state recognition is performed on the airborne remote sensing detection vector, the Z-th level flow state recognition is performed on the airborne remote sensing detection vector and the Z-1-th knowledge feature set generated by the Z-1-th time-series focusing subnet through the Z-th second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets, so as to obtain the Z-th level water area flow trend vector after the Z-th level flow state recognition of the airborne remote sensing detection data; Z is an integer greater than 1; The water area flow trend vector obtained after the last layer flow state identification of the pyramid flow state identification is determined as the multimodal sewage discharge monitoring vector obtained after the pyramid flow state identification.
2. The method according to claim 1, characterized in that: The plurality of second linear knowledge embedding subnets include: at least one second linear knowledge embedding subnet for performing feature compression on the airborne remote sensing detection vector, and at least one second linear knowledge embedding subnet for performing feature derivation on the airborne remote sensing detection vector; Among them, the number of second linear knowledge embedding subnets used for feature compression of the airborne remote sensing detection vector is the same as the number of second linear knowledge embedding subnets used for feature deriving of the airborne remote sensing detection vector; and the at least one second linear knowledge embedding subnet used for feature deriving of the airborne remote sensing detection vector is connected after the at least one second linear knowledge embedding subnet used for feature compression of the airborne remote sensing detection vector.
3. The method according to claim 1, characterized in that Before performing the first level flow state identification on the airborne remote sensing detection vector, the method further comprises: generating a water area flow trend disturbance vector of the flow state identification algorithm through a preconfigured disturbance processing branch; and loading the water area flow trend disturbance vector and the airborne remote sensing detection vector as inputs into the flow state identification algorithm; The method of performing first-layer flow state identification on the airborne remote sensing detection vector through the first second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets to obtain a first-layer water area flow trend vector of the airborne remote sensing detection data after the first-layer flow state identification includes: performing first-layer flow state identification on the airborne remote sensing detection vector and the water area flow trend disturbance vector through the first second linear knowledge embedding subnet to obtain the first-layer water area flow trend vector.
4. The method according to claim 1, characterized in that: The characteristic size of the flow trend vector of the Y-th layer of water area is consistent with the characteristic size of the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet; The Yth time-series focusing subnet in the flow state recognition algorithm performs the knowledge integration on the Yth layer of water area flow trend vector after the Yth layer of flow state recognition and the three-dimensional water body space inversion vector generated by the Yth first linear knowledge embedding subnet in the three-dimensional water body inversion algorithm to obtain the Yth knowledge feature set, including: Through the Yth time-series focusing subnet in the flow state recognition algorithm, the flow trend vector of the Yth layer of water area and the three-dimensional water body space inversion vector generated by the Yth first linear knowledge embedding subnet are feature fused at the local attention level to obtain a dynamic monitoring fusion vector; Performing feature extraction processing on the dynamic monitoring fusion vector to obtain a dynamic monitoring dense vector; Quantitative mapping is performed on the dynamic monitoring dense vector to obtain a confidence relationship network between the Y-th layer water flow trend vector and the three-dimensional water body space inversion vector generated by the Y-th first linear knowledge embedding subnet; Performing knowledge integration processing on the dynamic monitoring fusion vector according to the confidence relationship network to obtain the Yth knowledge feature set; The step of performing feature extraction processing on the dynamic monitoring fusion vector to obtain a dynamic monitoring dense vector includes: Performing local feature extraction processing and global feature extraction processing on the dynamic monitoring fusion vector respectively, and obtaining a local monitoring dense vector and a global monitoring dense vector accordingly; The local monitoring dense vector and the global monitoring dense vector are determined as the dynamic monitoring dense vector.
5. The method according to claim 1, characterized in that The feature mining is performed on the initial water quality sensor monitoring data and the airborne remote sensing detection data respectively, and the water quality sensor monitoring vector and the airborne remote sensing detection vector are obtained accordingly, including: Calling a remote sensing detection feature mining network to perform remote sensing detection feature mining on the airborne remote sensing detection data to obtain the airborne remote sensing detection vector; Calling a water quality monitoring feature mining network to perform water quality monitoring feature mining on the initial water quality sensor monitoring data to obtain the water quality sensor monitoring vector; The remote sensing dynamic monitoring method of the sewage outlet into the river is implemented by a pre-debugging deep learning network; the deep learning network includes a first deep learning branch and a second deep learning branch; the first deep learning branch includes at least a remote sensing detection feature mining network and a flow state recognition algorithm, and the second deep learning branch includes a water quality monitoring feature mining network and a three-dimensional water body inversion algorithm; the deep learning network is debugged through the following steps: Loading the water quality sensor monitoring data samples in the training sample set into the second deep learning branch, performing water quality monitoring feature mining on the water quality sensor monitoring data samples through the water quality monitoring feature mining network in the second deep learning branch to obtain water quality sensor monitoring vector samples; Performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector sample by using the three-dimensional water body inversion algorithm in the second deep learning branch, and obtaining three-dimensional water body space inversion vector samples of the monitoring task label samples in the water quality sensor monitoring data sample at several pyramid depths; Loading the airborne remote sensing detection data samples in the training sample set into the first deep learning branch, performing remote sensing detection feature mining on the airborne remote sensing detection data samples through the remote sensing detection feature mining network in the first deep learning branch, and obtaining airborne remote sensing detection vector samples; Performing pyramid flow state recognition on the airborne remote sensing detection vector samples in sequence through the flow state recognition algorithm in the first deep learning branch to obtain multi-modal pollution discharge monitoring vector samples; Determine the training error of the deep learning network based on the three-dimensional water body space inversion vector sample and the multimodal sewage monitoring vector sample to obtain the network training error of the deep learning network; The network weight variables in the deep learning network are improved according to the network training error to obtain a debugged deep learning network.
6. A remote sensing dynamic monitoring device for sewage outlets into rivers, characterized in that: The remote sensing dynamic monitoring device for the sewage outlet into the river is used for: Acquire initial water quality sensor monitoring data and airborne remote sensing detection data with target monitoring task labels; wherein the target monitoring task labels correspond to the target river discharge outlet area; Performing feature mining on the initial water quality sensor monitoring data and the airborne remote sensing detection data respectively, and obtaining a water quality sensor monitoring vector and an airborne remote sensing detection vector correspondingly; Performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and obtaining corresponding three-dimensional water body space inversion vectors of the target monitoring task label in the initial water quality sensor monitoring data at several pyramid depths; The airborne remote sensing detection vector is sequentially subjected to pyramid flow state recognition; wherein the pyramid depth of the pyramid flow state recognition is the same as the pyramid depth of the pyramid three-dimensional water body inversion; between each two consecutive layers of flow state recognition, the flow state recognition result obtained by the previous layer of flow state recognition in the two layers of flow state recognition and the three-dimensional water body space inversion vector corresponding to the pyramid depth are subjected to knowledge integration; when the latter layer of flow state recognition in the two layers of flow state recognition is subjected to flow state recognition on the airborne remote sensing detection vector and the knowledge feature set after the knowledge integration; The multimodal sewage monitoring vector obtained after the pyramid flow state recognition is processed to obtain linkage monitoring annotation data of the task attention vector associated with the target monitoring task label and the dynamic remote sensing detection trend vector of the airborne remote sensing detection data; Based on the obtained task attention vector associated with the target monitoring task label and the linkage monitoring annotation data of the dynamic remote sensing detection trend vector of the airborne remote sensing detection data, real-time monitoring and early warning analysis are performed on the target river discharge outlet area; wherein the task attention vector is used to highlight the data features most relevant to the target monitoring task, the data features are used to locate key water quality problems at the discharge outlet, and the dynamic remote sensing detection trend vector is used to provide a prediction of future water quality change trends, which is used to provide a reference for the early formulation of countermeasures to prevent water quality deterioration; The pyramid three-dimensional water body inversion is performed on the water quality sensor monitoring vector to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths, including: performing pyramid three-dimensional water body inversion on the water quality sensor monitoring vector through a plurality of feature inversion processing branches in a pre-debugged three-dimensional water body inversion algorithm to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths; wherein: when the first layer of three-dimensional water body inversion is performed on the water quality sensor monitoring vector, the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data is obtained through the plurality of feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm; The first feature inversion processing branch in the feature inversion processing branch performs the first layer of three-dimensional water body inversion on the water quality sensor monitoring vector to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at the first layer; when the water quality sensor monitoring vector is subjected to the Xth layer of three-dimensional water body inversion, the Xth feature inversion processing branch in the plurality of feature inversion processing branches performs the Xth layer of three-dimensional water body inversion on the three-dimensional water body space inversion vector of the X-1th layer to obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at the Xth layer; X is an integer greater than 1; The feature inversion processing branch is a first linear knowledge embedding subnet, and the first linear knowledge embedding subnet includes a linear knowledge embedding core; the plurality of feature inversion processing branches in the pre-debugged three-dimensional water body inversion algorithm perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and correspondingly obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths, including: by calling the linear knowledge embedding core in each first linear knowledge embedding subnet in the plurality of feature inversion processing branches, perform pyramid three-dimensional water body inversion on the water quality sensor monitoring vector, and correspondingly obtain the three-dimensional water body space inversion vector of the target monitoring task label in the initial water quality sensor monitoring data at a plurality of pyramid depths; The method sequentially performing pyramid flow state identification on the airborne remote sensing detection vector comprises: sequentially performing pyramid flow state identification on the airborne remote sensing detection vector through a pre-debugged flow state identification algorithm; wherein the flow state identification algorithm comprises a plurality of cascaded second linear knowledge embedding subnets, and a time-series focusing subnet is connected between each two consecutive second linear knowledge embedding subnets; the number of the second linear knowledge embedding subnets is the same as the number of the first linear knowledge embedding subnets in the three-dimensional water body inversion algorithm; The pre-debugged flow state recognition algorithm sequentially performs pyramid flow state recognition on the airborne remote sensing detection vector, including: When the airborne remote sensing detection vector is subjected to the first-layer flow state identification, the airborne remote sensing detection vector is subjected to the first-layer flow state identification through the first second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets, so as to obtain the first-layer water area flow trend vector of the airborne remote sensing detection data after the first-layer flow state identification; After performing the Yth layer flow state recognition on the airborne remote sensing detection vector, the Yth layer water area flow trend vector after the Yth layer flow state recognition and the three-dimensional water body space inversion vector generated by the Yth first linear knowledge embedding subnet in the three-dimensional water body inversion algorithm are subjected to the knowledge integration through the Yth time-series focusing subnet in the flow state recognition algorithm to obtain the Yth knowledge feature set; Y is an integer greater than 0; When the Z-th level flow state recognition is performed on the airborne remote sensing detection vector, the Z-th level flow state recognition is performed on the airborne remote sensing detection vector and the Z-1-th knowledge feature set generated by the Z-1-th time-series focusing subnet through the Z-th second linear knowledge embedding subnet among the plurality of second linear knowledge embedding subnets, so as to obtain the Z-th level water area flow trend vector after the Z-th level flow state recognition of the airborne remote sensing detection data; Z is an integer greater than 1; The water area flow trend vector obtained after the last layer flow state identification of the pyramid flow state identification is determined as the multimodal sewage discharge monitoring vector obtained after the pyramid flow state identification.
7. A remote sensing dynamic monitoring device, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 5.
8. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the method according to any one of claims 1 to 5 is implemented.
