Real-time identification and information transmission method and system for black and odorous water bodies in cities from space

Abstract

This application relates to the field of remote sensing image processing technology, specifically disclosing a method and system for real-time identification and information downlink of urban black and odorous water bodies from satellite, including: receiving and parsing task instructions; selecting images to be processed from image data stored on satellite according to the target area and observation time window in the task instructions; performing adaptive on-satellite processing on the images to be processed according to the processing parameters in the task instructions to obtain an identification result set including black and odorous water body identification results and pollution levels, floating garbage identification results, and green algae and duckweed identification results; scoring each identification result, with the score for black and odorous water body identification results calculated based on pollution level and area, and the scores for floating garbage identification results and green algae and duckweed identification results calculated based on area; sorting the scores of all identification results, selecting high-value identification results according to the sorting results, compressing the high-value identification results, and prioritizing their downlink to the ground via the satellite-to-ground data transmission channel.

Description

Technical Field

[0001] This application relates to the field of remote sensing image processing technology, and in particular to a method and system for real-time identification and information downlink of black and odorous water bodies in cities from space. Background Technology

[0002] With the acceleration of urbanization and the continuous improvement of environmental protection requirements in my country, there is an increasingly urgent need for routine and refined monitoring of urban water bodies, especially for the rapid detection and precise control of black and odorous water bodies that seriously affect the quality of life of residents and the ecological image of cities. Remote sensing technology, with its advantages of wide coverage, short revisit cycles, and lack of geographical limitations, has become an important means of urban water environment monitoring, particularly for large-scale black and odorous water bodies, and related technologies and methods continue to evolve.

[0003] Currently, the mainstream remote sensing identification and information acquisition model for urban black and odorous water bodies typically follows a technical path of "space-ground collaboration and ground processing." Specifically, the onboard computing platform first acquires multispectral or hyperspectral raw image data covering the target area and transmits this massive amount of raw data completely to the ground receiving station. Subsequently, on the ground server, threshold segmentation methods based on spectral indices (such as NDWI and BSI) or intelligent identification algorithms based on deep learning models (such as U-Net and SAM) are used to manually or semi-automatically process and analyze the transmitted images, ultimately extracting information such as the distribution and extent of black and odorous water bodies, and distributing the results to application departments. This technical system, after years of development, has achieved considerable advantages in terms of identification accuracy and algorithm maturity.

[0004] However, the existing technology model centered on ground processing has gradually revealed its inherent limitations when facing the practical needs of emergency monitoring and real-time control of urban water environments. First, there is a response delay of hours or even longer between satellite imagery and ground-based processing and results output, making it difficult to provide timely information support for precise governance with "immediate detection and immediate response." Sometimes, this even leads to the predicament where the water pollution status has changed by the time processing is completed. Second, the complete downloading of raw image data containing a large amount of invalid background information consumes extremely valuable satellite-to-ground data transmission bandwidth resources, resulting in high data transmission costs and low efficiency. Furthermore, the processing flow is heavily reliant on ground computing power and manual interpretation, and there is still room for improvement in automation and intelligence levels. Summary of the Invention

[0005] In view of this, this application provides a method and system for real-time identification and information downlink of urban black and odorous water bodies on satellite, as well as a storage medium and computer equipment. By receiving and parsing standardized ground commands to drive on-board processing, it realizes flexible definition and precise control of monitoring tasks. By directly executing a lightweight model on satellite for adaptive processing and identification, it generates an identification result set including black and odorous water bodies and their levels, floating garbage, and green algae and duckweed, realizing real-time conversion from raw data to effective information. Furthermore, it uses a value assessment model to uniformly quantify, score, and rank all identification results, and prioritizes the downlink of high-value information. Ultimately, it fundamentally overcomes the shortcomings of traditional methods, such as large response delays, high data transmission costs, and slow output of effective information, and achieves near real-time perception and efficient direct information delivery for environmental problems such as urban black and odorous water bodies.

[0006] According to one aspect of this application, a method for real-time identification and information downlink of black and odorous water bodies in cities from space is provided, including:

[0007] Receive and parse standardized mission instructions from the ground, wherein the mission instructions define the target area, observation time window and processing parameters;

[0008] Based on the target area and observation time window in the mission instructions, select spatiotemporally matching images to be processed from the image data stored on the satellite.

[0009] According to the processing parameters in the task instruction, adaptive on-board processing is performed on the image to be processed to obtain a set of identification results, including black and odorous water body identification results and corresponding pollution levels, floating garbage identification results, and green algae and duckweed identification results;

[0010] For each identification result in the identification result set, the identification result is scored using a value assessment model. The score for the black and odorous water body identification result is calculated based on the corresponding pollution level and area, while the scores for the floating garbage identification result and the green algae and duckweed identification result are calculated based on their respective corresponding areas.

[0011] The scores of all identification results are sorted, high-value identification results are selected based on the sorting results, the high-value identification results are compressed, and they are prioritized for downlink transmission to the ground via the satellite-to-ground data transmission channel.

[0012] According to another aspect of this application, a real-time identification and information transmission system for black and odorous water bodies in cities on space is provided, comprising:

[0013] The instruction receiving module is used to receive and parse standardized mission instructions from the ground, wherein the mission instructions define the target area, observation time window and processing parameters;

[0014] The image filtering module is used to filter out spatiotemporally matching images to be processed from the image data stored on the satellite according to the target area and observation time window in the mission instructions.

[0015] The identification module is used to perform adaptive on-board processing on the image to be processed according to the processing parameters in the task instruction, and obtain an identification result set including the identification results of black and odorous water bodies and the corresponding pollution levels, garbage floating identification results, and green algae and duckweed identification results;

[0016] The scoring calculation module is used to score each identification result in the identification result set using a value assessment model. The score of the black and odorous water body identification result is calculated based on the corresponding pollution level and area, and the scores of the floating garbage identification result and the green algae and duckweed identification result are calculated based on their respective areas.

[0017] The downlink module is used to sort the scores of all recognition results, select high-value recognition results based on the sorting results, compress the high-value recognition results, and prioritize their downlink to the ground via the satellite-to-ground data transmission channel.

[0018] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described method for real-time identification and information transmission of black and odorous water bodies in cities on satellites.

[0019] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor. When the processor executes the program, it implements the above-mentioned method for real-time identification and information transmission of black and odorous water bodies in cities on satellites.

[0020] Using the above technical solution, this application provides a method and system for real-time identification and information transmission of urban black and odorous water bodies on satellite, as well as storage media and computer equipment. First, when the satellite system is in orbit, it can receive and parse standardized mission instructions sent by the ground station via the satellite-to-ground data transmission channel to obtain the target area, the desired image observation time window, and specific processing parameters. Next, the satellite system can automatically search and match historical or real-time image databases stored internally, based on the target area and observation time window specified in the mission instructions, to select images to be processed. After obtaining the matched images to be processed, an adaptive on-board intelligent processing flow can be initiated according to the processing parameters carried in the mission instructions to generate a complete set of identification results, including black and odorous water bodies and their pollution levels, floating garbage, and green algae. Subsequently, each independent identification result in the set is quantitatively scored using a pre-set value assessment model. Specifically, for black and odorous water bodies, the score combines the severity coefficient of their pollution level with their area size; for floating garbage and green algae, the score is directly related to the area they cover. Finally, all identification results are uniformly sorted in descending order, and high-value identification results are selected based on the sorting results and preset quotas. These selected high-value identification results can be efficiently encapsulated using differentiated compression strategies, and are prioritized for transmission to the ground via limited satellite-to-ground data transmission channels based on their scores. This embodiment of the application achieves flexible definition and precise control of monitoring tasks by receiving and parsing standardized ground commands to drive onboard processing; by directly executing a lightweight model for adaptive processing and identification onboard, it generates an identification result set including black and odorous water bodies and their levels, floating garbage, and green algae and duckweed, achieving real-time conversion from raw data to effective information; furthermore, a value assessment model is used to uniformly quantify and sort all identification results, prioritizing the transmission of high-value information. Ultimately, this fundamentally overcomes the shortcomings of traditional methods, such as large response delays, high data transmission costs, and slow output of effective information, achieving near real-time perception and efficient direct information delivery for environmental problems such as urban black and odorous water bodies.

[0021] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0023] Figure 1 A flowchart illustrating a method for real-time identification and information transmission of black and odorous water bodies in satellite cities, provided in an embodiment of this application, is shown.

[0024] Figure 2 This paper illustrates a flowchart of another method for real-time identification and information transmission of black and odorous water bodies in cities on satellite, provided in an embodiment of this application.

[0025] Figure 3 A schematic diagram of the processing flow of an intelligent recognition processing module provided in an embodiment of this application is shown;

[0026] Figure 4 This illustration shows a schematic diagram of a real-time identification and information downlink system for urban black and odorous water bodies on satellite, provided in an embodiment of this application.

[0027] Figure 5 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0028] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0029] This embodiment provides a method for real-time identification and information transmission of black and odorous water bodies in cities from space, such as... Figure 1 As shown, the method includes:

[0030] Step 101: Receive and parse standardized mission instructions from the ground, wherein the mission instructions define the target area, observation time window, and processing parameters.

[0031] Step 102: Based on the target area and observation time window in the mission instructions, select spatiotemporally matching images to be processed from the image data stored on the satellite.

[0032] Step 103: According to the processing parameters in the task instruction, perform adaptive on-board processing on the image to be processed to obtain an identification result set including the black and odorous water body identification results and the corresponding pollution level, garbage floating identification results, and green algae and duckweed identification results.

[0033] Step 104: For each identification result in the identification result set, the identification result is scored using a value assessment model. The score of the black and odorous water body identification result is calculated based on the corresponding pollution level and area, and the scores of the floating garbage identification result and the green algae and duckweed identification result are calculated based on their respective areas.

[0034] Step 105: Sort the scores of all identification results, select high-value identification results based on the sorting results, compress the high-value identification results, and prioritize their transmission to the ground via the satellite-to-ground data transmission channel.

[0035] This application provides a method for real-time identification and information transmission of urban black and odorous water bodies from a satellite. First, while the satellite system is in orbit, it can receive and parse standardized task instructions sent by ground stations via a satellite-to-ground data transmission channel. Here, standardized task instructions refer to a structured data format, such as JSON, which includes at least the target area, the desired image observation time window, and specific processing parameters. This transforms the diverse monitoring needs of ground users into machine instructions that the satellite computing platform can accurately understand and execute, laying the foundation for subsequent automated processing. In a specific embodiment, a JSON standardized task instruction interface can be constructed to provide a unified description of task parameters, processing flow, and output requirements. Through task instruction parsing, a four-level processing pipeline including image screening, water body extraction, black and odorous water identification, and result compression is automatically generated. This supports dynamic adjustment of algorithm combinations based on satellite resource status, resolving the contradiction between fixed satellite processing flows and task diversity, and achieving precise satellite task management. Simultaneously, the data volume of a single task instruction is extremely small, significantly reducing the burden of satellite-to-ground transmission. Furthermore, it allows ground personnel to flexibly configure and dynamically adjust on-orbit processing. This mechanism avoids complex on-board software reconfiguration, greatly improving the practicality and on-orbit operation and maintenance efficiency of the on-board system. In a specific embodiment, processing parameters may include one of the following: the identifier and version number of the algorithm model to be invoked, the storage path of the model file, the decision threshold configuration of the decision tree classification model, the selection rules for the spectral index, the weight coefficients of the value assessment model, the selection of the result compression strategy, and the task priority. These parameters enable ground personnel to remotely and flexibly adjust the on-board processing flow, accuracy, and resource allocation strategies according to different monitoring needs (such as emergency response and routine patrols) and changing environmental conditions (such as different seasons or regions), without requiring complex on-orbit software reconfiguration of the satellite.

[0036] Next, the onboard system can automatically search and match historical or real-time images stored in its internal database based on the target area and observation time window specified in the mission instructions, thereby selecting images that were captured entirely within the observation time window and whose image coverage and geographical overlap with the target area meet the requirements.

[0037] After obtaining the matching image to be processed, an adaptive on-board intelligent processing flow can be initiated according to the processing parameters carried in the mission instructions. This flow generates a complete set of identification results, including three categories: black and odorous water bodies and their pollution levels, floating garbage, and green algae / duckweed. In a specific embodiment, this flow can perform pixel-level analysis of the image to be processed, simultaneously identifying and segmenting ordinary water body areas, floating garbage areas, and green algae / duckweed areas. Subsequently, for ordinary water body areas, the specific distribution and extent of black and odorous water bodies, as well as their corresponding pollution levels (e.g., mild or severe), can be further determined. It is important to note that for a given image to be processed, the same identification result can include multiple areas. For example, for the floating garbage identification result, specifically, the image to be processed can include multiple floating garbage identification areas, and each specific identification result, i.e., each specific identification area, is subsequently scored.

[0038] Subsequently, each individual identification result in the set of identification results is quantitatively scored using a pre-set value assessment model. Specifically, for black and odorous water bodies, the score combines the severity coefficient of their pollution level with their area size, calculated using a weighted formula to arrive at the final score; for floating garbage and green algae, the score is directly related to the area they cover—the larger the area, the higher the score. This design allows different types and natures of environmental problems to be compared on the same value scale.

[0039] Finally, all identification results are ranked in descending order, and based on the ranking and a preset quota, the highest-value results are selected to obtain high-value identification results. These selected high-value results are then efficiently encapsulated using differentiated compression strategies, significantly reducing data volume. Differential compression may include simplifying boundaries and selectively retaining attributes. During scheduled transmission, these high-value identification results will be prioritized for transmission to the ground via limited satellite-to-ground data transmission channels based on their scores, ensuring that the most critical environmental pollution information reaches users as quickly as possible.

[0040] By applying the technical solution of this embodiment, firstly, when the on-orbit system is running, it can receive and parse standardized mission instructions sent by the ground station through the satellite-to-ground data transmission channel to obtain the target area, the desired image observation time window, and specific processing parameters. Next, the on-orbit system can automatically search and match images in its internally stored historical or real-time image database based on the target area and observation time window specified in the mission instructions, thereby selecting images to be processed. After obtaining the matched images to be processed, it can initiate an adaptive on-orbit intelligent processing flow according to the processing parameters carried in the mission instructions, generating a complete set of identification results containing three categories: black and odorous water bodies and their pollution levels, floating garbage, and green algae. Subsequently, each independent identification result in the set is quantitatively scored using a pre-set value assessment model. Specifically, for black and odorous water bodies, the score combines the severity coefficient of their pollution level with their area size; for floating garbage and green algae, the score is directly related to the area they cover. Finally, all identification results are uniformly sorted in descending order, and high-value identification results are selected based on the sorting results and a preset quota. These selected high-value identification results can be efficiently encapsulated using differentiated compression strategies and prioritized for transmission to the ground via limited satellite-to-ground data transmission channels based on their high-value identification result scores. This application's embodiment achieves flexible definition and precise control of monitoring tasks by receiving and parsing standardized ground commands to drive onboard processing. By directly executing a lightweight model onboard for adaptive processing and identification, it generates a set of identification results including black and odorous water bodies and their levels, floating garbage, and green algae, achieving real-time transformation from raw data to effective information. Furthermore, a value assessment model is used to uniformly quantify, score, and rank all identification results, prioritizing the transmission of high-value information. Ultimately, this fundamentally overcomes the shortcomings of traditional methods, such as large response delays, high data transmission costs, and slow output of effective information, achieving near real-time perception and efficient direct information delivery for environmental problems such as urban black and odorous water bodies.

[0041] Optionally, in this embodiment, step 103, "performing adaptive on-board processing on the image to be processed to obtain a set of identification results including black and odorous water body identification results and corresponding pollution levels, floating garbage identification results, and green algae and duckweed identification results," includes: calling a preset lightweight image segmentation model to process the image to be processed, and simultaneously outputting the water body pixel mask, floating garbage pixel mask, and green algae and duckweed pixel mask in the image to be processed; and removing pixel regions that overlap with the floating garbage pixel mask and the green algae and duckweed pixel mask from the water body pixel mask. The process involves obtaining a mask for the exposed water body area; calculating the multispectral index of the target image area corresponding to the mask, and inputting the multispectral index into a pre-trained decision tree classification model for judgment, thereby obtaining the black and odorous water body identification result and the corresponding pollution level; determining the garbage floating identification result and the green algae floating identification result based on the garbage floating pixel mask and the green algae floating pixel mask, respectively; and merging the black and odorous water body identification result and the corresponding pollution level, the garbage floating identification result, and the green algae floating identification result into the identification result set.

[0042] In this embodiment, firstly, a lightweight image segmentation model pre-stored in the onboard computing platform is invoked to process the selected images to be processed. This lightweight image segmentation model can be a specially optimized deep learning network with significantly reduced computational complexity, enabling efficient operation under the limited computing power and memory resources of the onboard system. This model can simultaneously identify and segment different targets in the image to be processed, and after processing, it can simultaneously output three independent pixel masks—a binary image using black and white to identify the location of specific target pixels. Specifically, the model can generate three mask files, respectively marking all water pixels, all floating debris pixels, and all algae / duckweed pixels in the image to be processed.

[0043] Next, from the pixel mask representing all water bodies, pixel areas overlapping with the masks for floating debris and algae are precisely removed, thus obtaining the mask for exposed water areas—that is, the pure water surface area after removing interference from obvious debris or vegetation cover. Calculating the exposed water area mask is crucial because subsequent spectral analysis techniques for determining whether water is black and odorous rely on the spectral signal reflected by the water itself, and surface coverings severely pollute and distort this signal. Pre-removal ensures the purity of the data for subsequent analysis, a prerequisite for guaranteeing the accuracy of black and odorous water identification.

[0044] Subsequently, in-depth spectral feature analysis is performed on the target image region corresponding to the masked exposed water area. Specifically, a set of multispectral indices can be calculated for each pixel within the target image region. This is a key technology in the field of remote sensing, which uses mathematical formulas to combine the reflectance of different bands received by satellite sensors to quantify specific physicochemical properties of the water body, such as turbidity, chlorophyll concentration, and the content of colored soluble organic matter. These are all potential indicators of black and odorous water bodies. Afterward, these calculated multispectral index values ​​are input into a pre-trained decision tree classification model. The decision tree classification model is a machine learning model that mimics human judgment logic. It uses a series of "if...then..." rules to make layer-by-layer judgments on the input features, ultimately outputting a conclusion on whether the exposed water area belongs to a black and odorous water body and its pollution level (e.g., mild, moderate, severe).

[0045] Simultaneously, the pixel masks of the identified floating litter and duckweed are processed in parallel. Specifically, these two masks can be vectorized separately, converting patches composed of discrete pixels into polygonal vector patches defined by a sequence of boundary coordinate points, possessing attributes such as area and location. This process transforms the original imagery into a standard spatial data format that can be directly managed and analyzed in a Geographic Information System (GIS), thereby generating structured identification results for floating litter and duckweed. Here, the identification results for floating litter and duckweed can exist in the form of patches.

[0046] Finally, the black and odorous water body identification results and their pollution levels, the garbage floating identification results, and the green algae and duckweed identification results generated in the above steps are collected and integrated, and packaged together into a set of identification results with a clear structure and complete information.

[0047] This application embodiment achieves simultaneous identification of water bodies, garbage, and algae, as well as determination of black and odor levels, directly on the satellite by receiving ground commands. This fundamentally solves the problems of slow response and delayed extraction of effective information in traditional methods, and significantly improves the timeliness and efficiency of satellite-ground collaboration in urban water environment anomaly monitoring.

[0048] In this embodiment of the application, optionally, the step of "calling a preset lightweight image segmentation model to process the image to be processed, and simultaneously outputting the pixel masks of water bodies, floating garbage, and green algae in the image to be processed" includes: inputting the image to be processed into the encoder of the lightweight image segmentation model, simultaneously extracting and outputting multi-scale features of the image to be processed, wherein the multi-scale features include shallow texture features, mid-level semantic features, and deep abstract features; inputting the multi-scale features into the decoder of the lightweight image segmentation model, strengthening the regions representing real water bodies and attached objects in the deep abstract features and suppressing the regions representing typical interference in the deep abstract features through the spatial and channel dual attention module in the decoder, and generating an attention weight map based on the processing result; and multiplying the attention weight map and the deep abstract features element-wise through the product unit in the decoder. A spatially weighted deep guiding feature is obtained. The deep guiding feature is then fused with the mid-level semantic feature through the feature fusion unit in the decoder to obtain a primary fused feature. The mid-level semantic feature includes a feature response pattern sensitive to water body boundary morphology and spatial connectivity, and the primary fused feature is used to enhance this feature response pattern. The primary fused feature is then fused with the shallow texture feature through a feature selection gating mechanism in the decoder to obtain a target fused feature. During the fusion process, the feature selection gating mechanism adaptively selects and enhances feature channels related to spectral discrimination and feature channels related to micro-texture discrimination. The target fused feature is then input to three parallel prediction heads of the lightweight image segmentation model. These three parallel prediction heads simultaneously output the pixel masks for water bodies, floating garbage, and green algae / duckweed in the image to be processed.

[0049] In this embodiment, the image to be processed is first input into a specially designed lightweight image segmentation model. The first stage of this model is an encoder, which can simultaneously extract visual information at different levels from the image to be processed, i.e., multi-scale features. Specifically, these include: shallow texture features, such as fine textures like edges and color abrupt changes; intermediate semantic features, such as the outline of water bodies and the shape of large floating objects; and deep abstract features, such as the ability to determine whether a region belongs to water, land, or other land features as a whole. This step provides a complete information foundation from details to the global picture for subsequent fine segmentation.

[0050] Subsequently, these extracted multi-scale features can be input into the decoder part of the lightweight image segmentation model. The first stage in the decoder is a spatial and channel dual attention module. This module can intelligently analyze deep abstract features. Its working principle is as follows: on the one hand, it evaluates the importance of each location in the spatial dimension, enhancing the signals of areas that may correspond to real water bodies and surface attachments such as garbage and algae, and actively suppressing the signals of areas that represent typical interferences, such as building shadows, cloud shadows, and high-moisture soil; on the other hand, it selects which feature channels carry key information in the channel dimension. Finally, this module outputs an attention weight map, where the value of each location directly represents the probability that the pixel belongs to the target to be segmented.

[0051] The second stage in the decoder is the product unit. Next, the generated attention weight map is multiplied point-by-point with the original deep abstract features. This operation applies spatial attention to the deep abstract features. Features in regions with high weights, i.e., regions with a high probability of a target, are preserved or even enhanced, while features in regions with low weights, i.e., interfering regions, are significantly weakened. The final result is a spatially weighted deep guiding feature, which carries sanitized and emphasized high-level semantic guidance about where the target is located.

[0052] The third stage of the decoder is the feature fusion unit. Specifically, after obtaining the deep guiding features, these features can be fused with the mid-level semantic features from the encoder. The mid-level semantic features are highly sensitive to structural information such as the boundary morphology of water bodies and the spatial connectivity between different water body regions. Through fusion, the guidance provided by the deep guiding features for the target to be segmented is combined with the structural details provided by the mid-level semantic features to form primary fused features. This step significantly enhances the ability to accurately delineate water body boundaries and distinguish adjacent independent water bodies.

[0053] The fourth stage of the decoder is a feature selection gating mechanism. To enable the lightweight image segmentation model to distinguish between different targets on the water surface, such as clean water, garbage, and algae, the decoder further introduces a feature selection gating mechanism. This mechanism fuses the primary fusion features from the previous step with the texture features from the shallowest layer of the encoder. It's important to note that the fusion process is not a simple addition, but rather an adaptive selection and enhancement of two key information streams: one is feature channels sensitive to spectral differences, crucial for distinguishing the spectral features of black and odorous water; the other is feature channels sensitive to subtle texture differences, crucial for distinguishing smooth water surfaces, rough garbage, and granular algae. Through this dynamic selection, the final generated target fusion features possess a strong ability to determine both "what it is" and "where the boundary is." In one specific embodiment, the aforementioned adaptive selection process can be implemented using a lightweight gated neural network. This network can analyze the primary fusion features and shallow texture features to be fused and generate a weight vector in real time. Each value in this weight vector corresponds to the importance score of a feature channel: for channels that have been validated by the model in historical learning as being able to effectively capture spectral differences, such as channels corresponding to the ratio of near-infrared to red light, and channels that can effectively characterize subtle texture patterns, such as channels corresponding to local gray-level gradient changes, the gated neural network can assign them higher weights; conversely, for irrelevant or redundant channels, lower weights are assigned. In this way, in the subsequent weighted summation fusion, the feature channel signals with high weights can be significantly enhanced, while those with low weights are weakened.

[0054] Finally, the target fusion features rich in discriminative information are input in parallel into three structurally identical but task-independent prediction heads. Each prediction head is a lightweight computational layer, responsible for interpreting the general target fusion features into specific segmentation tasks: the first prediction head outputs the pixel locations of all water bodies (including clean and potentially polluted ones) in the image to be processed, i.e., the water body pixel mask; the second prediction head is responsible for locating the pixel mask of floating garbage; and the third prediction head outputs the pixel mask of algae and duckweed. Here, since each prediction head has an independent set of trainable parameters used to distinguish between water bodies, garbage, and algae during model training, it can also predict different targets during model application. Thus, the lightweight image segmentation model simultaneously completes accurate segmentation of three types of targets with a single forward computation.

[0055] This application embodiment efficiently extracts multi-scale features through an encoder, uses an attention mechanism to focus on key points and ignore interference, and then organically combines abstract semantic understanding, precise structural information, and subtle spectral texture discrimination through step-by-step feature fusion and intelligent gating selection. This enables the model to have powerful feature representation and multi-target discrimination capabilities while maintaining lightweight design. As a result, it can robustly and efficiently complete the synchronous and accurate extraction of water bodies and their attachments under the condition of strictly limited spaceborne resources, ensuring the real-time performance and reliability of the entire black and odorous water body identification process.

[0056] In this embodiment of the application, optionally, the step of "for the target image area corresponding to the mask of the exposed water body area, calculating the multispectral index of the target image area, and inputting the multispectral index into a pre-trained decision tree classification model for judgment, to obtain the black and odorous water body identification result and the corresponding pollution level" includes: obtaining the geographical location identifier of the target area and the seasonal identifier corresponding to the observation time window; obtaining the corresponding target spectral index set from a pre-set spectral index selection rule library on the satellite according to the geographical location identifier and the seasonal identifier, and using the spectral indexes included in the target spectral index set as the multispectral indexes of the target image area, wherein the spectral index selection rule library defines the most effective lightweight spectral index combination for identifying black and odorous water bodies under different regional and seasonal conditions; calculating the spectral index corresponding to each spectral index according to the target image area; and inputting all spectral indices corresponding to the target image area into a pre-set decision tree classification model with multi-level judgment nodes. The system sequentially judges each pixel in the target image region using the decision tree classification model, outputting the pollution level of the black and odorous water body. The judgment threshold of the decision tree classification model is pre-configured based on the background characteristics of the water body in the target image region and seasonal parameters. Based on the exposed water body region mask, spatially connected water body pixels are aggregated into independent candidate water body patches. For each candidate water body patch, the total area of ​​pixels identified as black and odorous water bodies within the candidate water body patch is counted. If the proportion of the total area in any candidate water body patch is lower than a preset threshold, then the candidate water body patch is determined as a normal water body patch, and pixels identified as black and odorous water bodies within the candidate water body patch are removed. Otherwise, the candidate water body patch is determined as a black and odorous water body patch. Based on the black and odorous water body patches, the black and odorous water body identification result is determined, and based on the black and odorous water body pollution level of each pixel identified as black and odorous water body within each candidate water body patch, the final pollution level of the black and odorous water body identification result is determined.

[0057] In this embodiment, before starting spectral analysis, two key environmental context information can be obtained: first, the geographical location of the target area, such as the North China Plain or the Yangtze River Delta region; and second, the seasonal identifier calculated based on the observation time window, such as summer or winter. This allows the application of analysis rules that best match the current scenario, as the spectral characteristics of water bodies vary significantly with regional water body types (such as reservoirs and rivers) and seasons (temperature, light intensity, and aquatic vegetation cycles).

[0058] Next, using the aforementioned geographic location and seasonal identifiers as query keys, a pre-built spectral index selection rule base on the satellite is accessed. This rule base is essentially an expert knowledge base, encapsulating lightweight spectral index combinations that have been validated through extensive ground-based experiments under different regional and seasonal conditions, and are most effective in identifying black and odorous water bodies. Then, the corresponding target spectral index set can be retrieved from the spectral index selection rule base. For example, in the case of southern urban rivers during summer, this set could include the CDOM index, which focuses on reflecting organic matter concentration, and specific band ratios reflecting chlorophyll a. This ensures that subsequent calculations focus on the most discriminative features, avoiding computational redundancy.

[0059] Then, for the target image area corresponding to the mask of the exposed water body area, based on the target spectral index set selected in the previous step, the specific value corresponding to each spectral index is calculated pixel by pixel, that is, the spectral index. For example, the NDWI (Normalized Differential Water Index) value and BSI (Black and Odorous Water Index) value of each water body pixel are calculated.

[0060] Subsequently, the spectral indices of all pixels are input into a pre-trained decision tree classification model. This model consists of multi-level decision nodes, which can perform layer-by-layer judgments on multiple spectral indices of each pixel according to pre-set threshold rules configured based on the background characteristics of the target area's water bodies and seasonal parameters. For example, if spectral index A is greater than threshold X, it proceeds to decision branch B; otherwise, it proceeds to branch C. The background characteristics of the target area's water bodies specifically refer to the spectral representation benchmark of the inherent optical and physicochemical properties of typical water bodies in the target area under normal conditions without black and odorous pollution, as shown in remote sensing images. After a series of judgments, the decision tree classification model can output a preliminary black and odorous water pollution level for each water body pixel, such as normal, slightly black and odorous, or severely black and odorous. This completes the initial pixel-level classification.

[0061] After obtaining the initial pixel-level classification results, geographically significant objects can be constructed based on the pixels. Specifically, based on the initial mask of the exposed water body area, region growing or connected component analysis algorithms in image processing can be used to aggregate spatially adjacent water body pixels to form independent candidate water body patches. Each candidate water body patch represents a physically continuous water body area, such as a pond or a section of a river.

[0062] Next, internal statistics are performed on each candidate water body patch. Specifically, all pixels within the candidate water body patch can be traversed, and based on the preliminary judgment of the decision tree classification model in the previous step, the total area of ​​all pixels marked as black and odorous (whether mild or severe) is accumulated. This total area can be used to measure the spatial influence range of the black and odorous phenomenon within the candidate water body patch.

[0063] Then, for any candidate water body patch, if the total area of ​​the black and odorous pixels within it accounts for less than a preset confidence threshold, these scattered black and odorous pixels are considered likely misjudged, perhaps due to accidental factors such as shadows or wave surges. The entire candidate water body patch is then classified as a normal water body patch. To ensure a clean result, all pixels initially identified as black and odorous within these candidate water body patches can be reclassified as normal. Conversely, if the area of ​​black and odorous pixels reaches or exceeds the confidence threshold, the candidate water body patch is confirmed as a black and odorous water body patch.

[0064] Finally, for all objects identified as black and odorous water bodies, their geographical range and attributes are packaged to determine the final black and odorous water body identification result. Simultaneously, the pollution level distribution of all pixels identified as black and odorous within the patch can be analyzed to determine and output the final pollution level representing the overall pollution degree of the patch. For example, the pollution level with the highest percentage can be selected as the final pollution level of the overall pollution degree of the patch.

[0065] This application's embodiments dynamically select the optimal spectral index based on geographical location and seasonal indicators, significantly improving the discrimination accuracy of the decision tree classification model in different scenarios. Furthermore, the decision tree classification model has a clear structure and high computational efficiency, making it highly suitable for satellite deployment. Additionally, this application's embodiments introduce a post-processing step based on confidence level verification of patch area proportions. This effectively filters out unavoidable scattered noise and misclassifications in pixel-level classification, aggregating scattered suspected pixels into contaminated patches with decision-making significance, thereby greatly improving the reliability and practicality of the final identification results.

[0066] Optionally, in this embodiment of the application, step 102 includes:

[0067] Step 102-1: Based on the observation time window in the mission instruction, select candidate images from the image data stored on the satellite whose image acquisition time falls entirely within the observation time window.

[0068] Step 102-2: For each candidate image, calculate the spatial overlap between the geographic coverage of the candidate image and the target area in the task instruction;

[0069] Step 102-3: Candidate images with spatial overlap greater than or equal to a preset overlap threshold are determined as spatiotemporally matched images to be processed.

[0070] In this embodiment, firstly, a search is performed in the satellite's storage system according to the observation time window explicitly specified in the mission instructions. Here, the observation time window refers to a specific time period required by the mission instructions sent from the ground, such as "9:00 AM to 10:00 AM on October 1, 2023". Next, the acquisition time of each stored image can be checked, i.e., the moment the satellite actually captured the image, and images whose start and end times fall entirely within the observation time window are selected and marked as candidate images. This ensures that the data used in subsequent processing is highly consistent with the user's monitoring needs in the time dimension.

[0071] Next, a fine-grained calculation of spatial matching degree is performed on each selected candidate image. Each candidate image corresponds to a specific geographic coverage area, that is, the area it covers on the actual earth's surface. Specifically, geospatial calculations are used to determine the area of ​​intersection between the geographic coverage area of ​​the candidate image and the target area defined in the task instructions, and the proportion of this intersection area to the total area of ​​the target area is calculated. This proportion is the spatial overlap degree. This calculation quantifies the spatial relevance of the candidate image to the target monitoring task; the higher the spatial overlap degree, the more complete the target area information contained in the candidate image.

[0072] Finally, the calculated spatial overlap is compared with a preset overlap threshold. This preset overlap threshold can be an empirical value, such as 75%. Only when the spatial overlap of a candidate image is greater than or equal to this threshold is it ultimately determined to be a spatiotemporally matched image to be processed. This mechanism ensures that the selected candidate images not only meet the temporal requirements, but also that their main content actually covers the core area of ​​interest to the user, effectively filtering out candidate images that only scan the edge of the target area or do not fully cover it, thus ensuring that subsequent processing algorithms can obtain high-quality, highly relevant input data.

[0073] This application embodiment, through strict spatiotemporal constraints, accurately and efficiently locates a small number of images most relevant to the current task from massive on-board storage, greatly reducing the ineffective consumption of subsequent expensive intelligent computing resources and laying the foundation for the smoothness and efficiency of the entire on-board real-time processing flow.

[0074] In this embodiment of the application, optionally, the processing parameters include task priority; before "performing adaptive on-board processing on the image to be processed" in step 103, the method further includes: real-time monitoring of the current comprehensive resource load status, wherein the comprehensive resource load status is determined based on at least one of the current available memory, CPU utilization, remaining power budget, and the expected computational complexity of the current image to be processed; comparing the comprehensive resource load status with multiple preset resource level thresholds to determine the target resource level, and optimizing the configuration of the lightweight image segmentation model and the decision tree classification model according to the target resource level and the task priority defined in the task instruction; wherein, when the target resource level indicates that the current resource load is light... When the resource scarcity level is normal and the task priority is normal, the first level of optimization is performed, which involves pausing the calculation of the redundancy spectral index in the decision tree classification model. When the target resource level indicates a moderate resource scarcity level, or the task priority is high, the first and second levels of optimization are performed, where the second level of optimization dynamically skips the preset non-critical feature extraction layer in the lightweight image segmentation model. When the target resource level indicates a severe resource scarcity level, the first, second, and third levels of optimization are performed, where the third level of optimization determines high-risk water bodies in the image to be processed based on prior urban geographic information, and adaptive on-board processing is performed only on the high-risk water bodies.

[0075] In this embodiment, a resource monitoring and evaluation process can be initiated before processing the image to be processed. Specifically, multi-dimensional operational indicators of the onboard computing platform can be collected in real time, and the overall resource load status can be determined based on these indicators. Here, multi-dimensional operational indicators may include: the amount of currently available free memory, which directly limits the scale of data and models that can be loaded simultaneously; the utilization rate of the central processing unit (CPU), reflecting the level of computing power; the remaining power budget allocated by the onboard computing platform to the current task, which is the energy basis for all operations on the satellite; and the estimated computational complexity assessment of the image to be processed, such as estimating its processing difficulty based on image size, number of bands, and texture richness. By integrating these dynamic and static indicators, the busy status of the onboard computing platform can be accurately perceived.

[0076] Next, the overall resource load status is compared with a series of preset resource level thresholds. Based on these thresholds, the resource status can be categorized into multiple levels, such as relaxed, slightly strained, moderately strained, and severely strained. After determining the current target resource level, it can be combined with the task priorities specified in the ground mission instructions for joint decision-making. Task priorities can be, for example, routine patrols or high-priority emergency response. Based on this decision, the lightweight image segmentation model and decision tree classification model can be dynamically optimized, i.e., their operating methods and resource allocation strategies can be adjusted to seek the optimal result under limited conditions.

[0077] The optimization strategy can be implemented in stages based on resource status and task priority. When the target resource level reflects a current resource shortage level and the task priority is normal, the first level of optimization can be performed, which involves pausing the calculation of redundant spectral indices in the decision tree classification model. Here, redundant spectral indices refer to derived indices that contribute relatively little to the final black and odorous substance determination but have high computational costs. Pausing the calculation of these redundant spectral indices can quickly reduce the computational load with almost no impact on the core conclusions.

[0078] When the target resource level reflects a moderate resource shortage or a high task priority, a second level of optimization can be added on top of the first level. This optimization applies to lightweight image segmentation models by dynamically skipping pre-defined non-critical feature extraction layers in the network. Deep learning models typically consist of multiple layers, where some intermediate layers have a relatively minor impact on the final segmentation accuracy. When resources are limited, these layers can be intelligently bypassed, sacrificing a small amount of accuracy for a significant increase in processing speed, ensuring the task can be completed within the required time.

[0079] When the target resource level reflects a severe resource shortage, the highest level of optimization (Level 3) can be initiated to ensure the core mission is not interrupted. This optimization incorporates prior urban geographic information, namely a pre-installed spatial database on the satellite containing key locations such as urban built-up areas, major industrial zones, and known sewage outlets. Using this knowledge, the water bodies with the highest pollution risk are quickly identified in the imagery to be processed. Subsequently, all valuable computational resources are focused on identifying and processing these high-risk water bodies, while other low-risk areas are temporarily ignored. This ensures that the most valuable information is successfully acquired even under extreme conditions.

[0080] The embodiments of this application can dynamically and progressively adjust the algorithm complexity and processing range according to real-time resource availability, task urgency and data characteristics. This design enables the entire onboard computing platform to have excellent resilience and robustness in the space environment where computing, memory and power consumption are extremely limited. It can always keep running under various pressure conditions and prioritize the output of the most critical tasks, achieving a high degree of unity between intelligence, efficiency and reliability.

[0081] Optionally, in this embodiment of the application, the value assessment model includes a black and odorous water body scoring sub-model for scoring the identification results of the black and odorous water body; the black and odorous water body scoring sub-model is as follows:

[0082] S = α × G + β × log(A);

[0083] Where S represents the score of the black and odorous water body identification result, G represents the pollution level, A represents the area of ​​the black and odorous water body identification result, and α and β are configurable weight parameters.

[0084] In this embodiment, the above formula can comprehensively measure the severity and spatial impact of black and odorous water body incidents. The first part of the formula, α×G, evaluates the severity of the pollution. G is a numerical value corresponding to the pollution level; for example, 1 corresponds to mild, 2 to moderate, and 3 to severe, with higher pollution levels resulting in larger values. α is an adjustable weighting parameter that controls the proportion of pollution severity in the final score. In scenarios requiring rapid response to emergency pollution events, the value of α can be increased to make the results more focused on the pollution level.

[0085] The second part of the formula, β×log(A), is used to evaluate the spatial scale of the pollution's impact. Here, A represents the actual area of ​​the identified black and odorous water body. Notably, the area A is not used directly; instead, its logarithm, log(A), is taken. This prevents a single, extremely large polluted water body from dominating the score, thus obscuring smaller but critical targets with extremely high pollution levels requiring urgent treatment. β is another weighting parameter paired with α, used to adjust the influence of the area factor.

[0086] This application embodiment uses a logarithmic function to process the area, effectively balancing the relationship between pollution severity and spatial scale, avoiding distortion from a single indicator; while the configurability of the weight parameters α and β gives ground control personnel the ability to flexibly adjust the scoring strategy according to task priority, ensuring that the limited communication resources on the satellite are always prioritized for transmitting the most valuable environmental pollution information, greatly improving the decision-making intelligence and practical efficiency of the entire monitoring system.

[0087] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the specific implementation process of this embodiment, another method for real-time identification and information transmission of urban black and odorous water bodies on satellite is provided, such as... Figure 2 As shown, the method includes:

[0088] First, ground monitoring personnel generate a structured task instruction for monitoring black and odorous water bodies at the ground station based on specific environmental monitoring needs. This task instruction can use standardized formats such as JSON and clearly includes parameters such as the geographic coordinates of the target area, the desired observation time window, imaging quality requirements such as cloud cover, and task processing priorities. After generation, the instruction is uploaded to the satellite platform in orbit via a dedicated telemetry and control link.

[0089] Once the mission instructions arrive at the satellite, they are received by the instruction parsing and mission scheduling module in the onboard computer (i.e., the onboard computing platform). This module first performs syntax and security checks on the instructions to ensure their correct format and reliable origin. Then, it precisely parses out the various parameters in the instructions and, based on these parameters and the satellite's current resource status, generates a specific mission execution plan, scheduling subsequent modules to work in an orderly manner. This module is crucial to the overall system's flexibility, enabling the satellite to dynamically respond to different missions, rather than simply operating according to a fixed program.

[0090] Next, the data management and preprocessing module begins operation. Based on the target area and observation time window provided by the instruction parsing and task scheduling module, it performs a rapid search within the image pool of historically or currently captured remote sensing images. Its core function is spatiotemporal matching, precisely identifying raw images whose capture time meets the requirements and whose geographical coverage highly overlaps with the target area. These selected images are then prepared and sent to the next processing stage.

[0091] Subsequently, the intelligent recognition and processing module is activated. This module calls upon model files stored on the satellite, which contain image segmentation models and decision tree classification models trained on the ground and optimized with deep lightweight technology. It can also reference information from a prior knowledge base, such as city boundaries and industrial area locations. The module first uses the lightweight image segmentation model to extract all water bodies and surface deposits, such as garbage and algae, from the image to be processed. Then, it calculates multispectral indices specifically for exposed water bodies and combines this with the decision tree classification model to determine whether the water is black and smelly and its pollution level.

[0092] Then, the results generation and downlink module performs value aggregation and refinement on all identified results, such as black and odorous water bodies, garbage, and green algae. It can use a value assessment model to calculate a score for each result, and then uniformly sort and filter them based on the scores, retaining only the highest-value results. Afterwards, these high-value results are differentially compressed, greatly reducing data volume. Finally, the processed, concise, and important downlink information is packaged and prioritized according to its value, then sent back to the satellite's data transmission antenna and transmitted to the ground station.

[0093] This application's embodiments move the cumbersome processing workflow, which traditionally relies on large ground servers and manual interpretation, to the satellite platform. Through a series of lightweight, modular intelligent processing units, the transformation and filtering from raw images to effective information are completed directly at the source of data generation. This fundamentally solves the three major pain points of traditional methods: high data downlink pressure, delayed information extraction, and slow response speed. It achieves near real-time perception and efficient information delivery for environmental problems such as urban black and odorous water bodies, significantly improving the utilization efficiency of satellite resources and the ability to respond to environmental emergencies.

[0094] In one specific embodiment Figure 2 The intelligent recognition processing module can include, for example, Figure 3 The system includes a water body intelligent identification module and a black and odorous water body identification module. Here, the water body intelligent identification module can embed the aforementioned lightweight image segmentation model, and the black and odorous water body identification module can embed the aforementioned decision tree classification model. The slice data can be the aforementioned image to be processed, and the water body slice data can be the aforementioned mask of exposed water body areas.

[0095] like Figure 3 As shown, the intelligent water body recognition module first processes the input slice data. This module's built-in lightweight image segmentation model can perform pixel-level fine classification of the image to be processed, identifying and categorizing pixels into: water bodies indicating floating garbage covered by solid waste, water bodies indicating algae or duckweed growth, and other relatively clean water bodies without obvious attachments. This completes the rapid identification and classification of visible pollution or abnormal coverage on the water surface.

[0096] Meanwhile, the black and odorous water body identification module specifically targets other water bodies—those exposed water areas not covered by garbage or algae—for water quality diagnosis. Instead of analyzing image texture, it calculates the multispectral indices of pixels in these areas. This involves using quantitative indicators derived from the spectral reflectance characteristics of water bodies in different wavelength bands to infer their inherent chemical and biological parameters. These indices are then input into a decision tree classification model. Based on pre-set threshold rules that consider seasonal and regional adaptability, the model determines whether the water body is black and odorous and the severity of the pollution. Finally, it outputs the black and odorous water body identification result and the corresponding pollution level, thus achieving invisible spectral detection of internal pollution in water bodies.

[0097] Subsequently, the results of garbage floating and green algae identification from the water body intelligent identification module, as well as the black and odorous water body classification results from the black and odorous water body identification module, are sent together to the key information packaging stage for problematic water bodies. In this stage, the core attributes of each identification result can be extracted and encapsulated, such as geographical coordinates, outline boundaries, area size, problem type, and pollution level, forming a structured data package.

[0098] Finally, all packaged structured information is transferred to the downlink preparation phase. During this phase, based on the communication link status and the urgency of the information, these data packets can be scheduled to be transmitted back to the ground receiving station via the satellite's data transmission antenna. At this point, the onboard processing is complete, and the ground receiving station will directly receive a clear and concise report of the environmental incident, rather than the raw imagery that requires further processing.

[0099] In one specific embodiment, after "receiving and parsing standardized mission instructions from the ground" in step 101, the method further includes:

[0100] According to the hierarchical parameter configuration system in the task instruction, the task-level global strategy, model-level algorithm configuration and resource-level constraint parameters are loaded, and the lightweight image segmentation model and decision tree classification model are initialized according to the task-level global strategy, the model-level algorithm configuration and the resource-level constraint parameters. The model-level algorithm configuration includes one or more spectral index discrimination thresholds for the decision tree classification model to determine black and odorous water bodies.

[0101] While the lightweight image segmentation model is used to process the image to be processed, the overall resource load status of the satellite is monitored in real time. The monitoring results are compared with the resource-level constraint parameters. If the comparison results indicate that resources are tight, the optimization configuration is dynamically triggered. The optimization configuration includes pausing the calculation of the spectral index marked as redundant in the model-level algorithm configuration in the decision tree classification model, and / or dynamically skipping the preset non-critical feature extraction layer in the lightweight image segmentation model.

[0102] After obtaining the preliminary black and odorous water body identification result of the image to be processed based on the spectral index discrimination threshold through the decision tree classification model, the preliminary black and odorous water body identification result is statistically analyzed. If the proportion of pixel area identified as black and odorous water body exceeds the preset abnormal threshold within any preset spatial range, it is determined that the current spectral index discrimination threshold is not applicable, and the spectral index discrimination threshold is automatically tightened and fine-tuned based on the adjustment strategy in the model-level algorithm configuration.

[0103] By using the finely adjusted spectral index to determine the threshold, the current image to be processed is reprocessed to obtain the final recognition result set.

[0104] In this embodiment, firstly, standardized task instructions from the ground are parsed, and subsequent processes are precisely set according to the preset hierarchical parameter configuration system. Specifically, the task-level global strategy defines the macro-level objectives of the task, such as whether a rapid response or extremely high accuracy is required, or whether it is a high-priority or low-priority task. The model-level algorithm configuration includes various algorithm parameters required to perform specific identification tasks, especially the spectral index discrimination threshold used by the decision tree classification model to distinguish whether water bodies are black and odorous. This threshold is a numerical standard derived from the analysis of a large amount of ground data. When the spectral characteristics reflected by the water body exceed a specific spectral index discrimination threshold, it is judged as potentially black and odorous. The resource-level constraint parameters clarify the upper limit of on-board computing resources that can be used for this task. After loading these parameters, the lightweight image segmentation model and the decision tree classification model are initialized accordingly, ensuring that there is clear strategy guidance and resource boundaries from the start of the task.

[0105] After initialization, the parallel processing and monitoring phase begins. While the lightweight image segmentation model performs pixel-level analysis of the image to be processed to distinguish between water, garbage, and algae, an independent monitoring unit tracks the overall resource load status of the onboard platform in real time. This includes key indicators such as currently available computing memory, CPU utilization, and remaining energy budget. This real-time monitoring data is continuously compared with the resource-level constraints preset in the task instructions. Once the comparison results indicate resource scarcity, such as impending memory exhaustion, dynamic optimization can be triggered immediately. This is an intelligent degradation processing strategy. For example, in the decision tree classification model, the computation of spectral indices marked as redundant by the model-level algorithm configuration can be paused (these indices contribute little to the final judgment but have high computational costs), and / or certain preset non-critical feature extraction layers can be dynamically skipped in the lightweight image segmentation model (these layers are used to extract more refined image features but can be temporarily omitted to improve speed when resources are insufficient). This ensures that the task can continue to run stably under limited resource conditions while guaranteeing the core recognition function.

[0106] When a decision tree classification model uses the spectral index discrimination threshold configured in the model-level algorithm to analyze exposed water bodies and outputs preliminary black and odorous water body identification results, it does not directly adopt these results. Instead, it initiates a self-checking and feedback loop. Specifically, it can perform statistical analysis on these preliminary black and odorous water body identification results, checking the proportion of the total area of ​​pixels identified as black and odorous water bodies to the total area of ​​water bodies in any preset geographic space (e.g., an administrative block or an independent water body patch). If this proportion exceeds a preset abnormal threshold (e.g., 50%), this abnormally high proportion often means that the currently used spectral index discrimination threshold may be set too leniently, leading to a large number of normal water bodies being misidentified, or encountering special interference scenarios. In this case, it automatically determines that the current spectral index discrimination threshold is inapplicable and, according to the preset adjustment strategy in the model-level algorithm configuration, performs a fine-tuning of the relevant spectral index discrimination threshold, i.e., raising the judgment standard, making subsequent identification more rigorous and accurate.

[0107] Finally, the image to be processed is reprocessed using a finely tuned and more reliable spectral index discrimination threshold. This step effectively corrects misjudgments caused by initial parameter mismatch, thereby generating a more accurate and reliable final identification result set. This result set not only includes verified and classified information on black and odorous water bodies, but also integrates information on floating garbage and green algae identified synchronously by a lightweight image segmentation model, together forming a complete water environment problem diagnostic report.

[0108] This application's embodiments achieve refined task management through a hierarchical parameter configuration system, ensure operational resilience in extreme onboard environments through resource awareness and dynamic degradation, and significantly improve the scene adaptability of the identification algorithm and the accuracy of the final results through a threshold self-adjustment mechanism based on statistical results. Thus, efficient and reliable intelligent monitoring of urban water environment is achieved under the condition of strictly limited onboard resources.

[0109] In another specific embodiment, the "compressing the high-value identification result" in step 103 includes:

[0110] The high-value identification results are categorized into black and odorous water body identification results, floating garbage identification results, and green algae and duckweed identification results.

[0111] For the identification results of floating garbage and the identification results of green algae and duckweed, core attributes are encapsulated for each identification result, and a lightweight structured data file is generated based on the encapsulated core attributes. The core attributes include a unique identifier, spatial coordinates representing geographical location, and area of ​​the map patch.

[0112] For the black and odorous water body identification results, a vector boundary data file describing the precise spatial range, a thematic map data file for visualization, and a statistical table data file recording detailed attributes are generated for each black and odorous water body patch in the black and odorous water body identification results.

[0113] The value score of each black and odorous water body patch is calculated, and all black and odorous water body patches are divided into multiple importance levels based on the value score.

[0114] For each black and odorous water body patch, based on the importance level corresponding to the black and odorous water body patch, the vector boundary data file, thematic map data file, and statistical table data file corresponding to the black and odorous water body patch are simplified and compressed.

[0115] In this embodiment, the high-value identification results are first categorized, forming the basis for subsequent differentiated processing. For surface attachment issues such as floating garbage and algae, a highly simplified information encapsulation strategy is implemented. Specifically, a set of core attributes is extracted and encapsulated for each identification result. These core attributes may include an ID for unique identification, spatial coordinates for geographic location (typically the latitude and longitude of the centroid), and patch area representing its size. Based on these core attributes, a lightweight structured data file (such as JSON format) is generated. This format has a clear structure and extremely small data volume; its fundamental purpose is to minimize data volume while meeting basic positioning and statistical requirements. That is, the structured data file is subsequently transmitted to the ground.

[0116] For the more critical identification results of black and odorous water bodies, a complete and multi-faceted data preservation strategy is adopted. Specifically, three types of data files are generated simultaneously for each black and odorous water body patch: First, vector boundary data files (such as SHP format), which is a standard format in Geographic Information Systems (GIS) used to accurately record the spatial location and shape of points, lines, and polygons, ensuring the geometric accuracy of water body boundaries; second, thematic map data files, which are visualization images generated by combining the identification results with the geographic base map and rendering them according to a specific style (such as using different colors to represent different pollution levels), facilitating an intuitive and rapid understanding of the spatial distribution of pollution; and third, statistical table data files, which record in detail the multi-dimensional attributes of each patch, such as its number, center location, pollution level, and area, in a structured table format (such as JSON or CSV), providing a data foundation for in-depth analysis.

[0117] Next, using a pre-defined value assessment model (e.g., comprehensively considering pollution level and area of ​​black and odorous water bodies), a quantitative value score is calculated for each black and odorous water body patch. Based on this value score, all black and odorous water body patches are divided into multiple importance levels (e.g., high, medium, and low). Subsequently, based on the importance level of the black and odorous water body patch, differentiated compression processing is applied to its corresponding three types of data files. For high-value black and odorous water body patches, low-loss or lossless compression is typically used for all three types of data files to ensure information integrity; for medium-value black and odorous water body patches, the vector boundary data files can be moderately simplified, and thematic map data files can be optimized, while retaining complete attributes; for low-value black and odorous water body patches, greater geometric simplification is allowed, thereby achieving significant data reduction.

[0118] In summary, the embodiments of this application avoid transmitting redundant data for secondary targets through a differentiated encapsulation strategy; and significantly reduce the amount of low-value data while ensuring the integrity of high-value information through a hierarchical compression mechanism. This refined processing increases the proportion of effective information at the source and greatly optimizes the utilization efficiency of valuable satellite-to-ground data transmission bandwidth.

[0119] Furthermore, as Figure 1 In terms of specific implementation, this application provides a real-time identification and information transmission system for urban black and odorous water bodies on satellite, such as... Figure 4 As shown, the system includes:

[0120] The instruction receiving module is used to receive and parse standardized mission instructions from the ground, wherein the mission instructions define the target area, observation time window and processing parameters;

[0121] The image filtering module is used to filter out spatiotemporally matching images to be processed from the image data stored on the satellite according to the target area and observation time window in the mission instructions.

[0122] The identification module is used to perform adaptive on-board processing on the image to be processed according to the processing parameters in the task instruction, and obtain an identification result set including the identification results of black and odorous water bodies and the corresponding pollution levels, garbage floating identification results, and green algae and duckweed identification results;

[0123] The scoring calculation module is used to score each identification result in the identification result set using a value assessment model. The score of the black and odorous water body identification result is calculated based on the corresponding pollution level and area, and the scores of the floating garbage identification result and the green algae and duckweed identification result are calculated based on their respective areas.

[0124] The downlink module is used to sort the scores of all recognition results, select high-value recognition results based on the sorting results, compress the high-value recognition results, and prioritize their downlink to the ground via the satellite-to-ground data transmission channel.

[0125] Optionally, the identification module is used for:

[0126] The preset lightweight image segmentation model is called to process the image to be processed, and the pixel masks of water bodies, floating garbage, and green algae and duckweed in the image to be processed are output simultaneously.

[0127] From the water body pixel mask, remove the pixel areas that overlap with the garbage floating pixel mask and the green algae and duckweed pixel mask to obtain the exposed water body area mask;

[0128] For the target image region corresponding to the mask of the exposed water body area, calculate the multispectral index of the target image region, and input the multispectral index into the pre-trained decision tree classification model for judgment to obtain the black and odorous water body identification result and the corresponding pollution level;

[0129] The garbage floating identification result and the green algae and duckweed identification result are determined respectively based on the pixel mask of the floating garbage and the pixel mask of the green algae and duckweed;

[0130] The identification results of the black and odorous water bodies, along with their corresponding pollution levels, the identification results of the floating garbage, and the identification results of the green algae and duckweed, are combined into the identification result set.

[0131] Optionally, the identification module is further configured to:

[0132] The image to be processed is input into the encoder of the lightweight image segmentation model, and the multi-scale features of the image to be processed are extracted and output simultaneously. The multi-scale features include shallow texture features, mid-level semantic features and deep abstract features.

[0133] The multi-scale features are input into the decoder of the lightweight image segmentation model. Through the spatial and channel dual attention module in the decoder, the regions representing real water bodies and attachments in the deep abstract features are strengthened, while the regions representing typical interferences in the deep abstract features are suppressed. An attention weight map is generated based on the processing results.

[0134] The attention weight map and the deep abstract features are multiplied element-wise by the product unit in the decoder to obtain spatially weighted deep guided features.

[0135] The deep guiding features and the mid-level semantic features are fused together by the feature fusion unit in the decoder to obtain primary fused features. The mid-level semantic features include feature response patterns that are sensitive to the morphology of water body boundaries and spatial connectivity. The primary fused features are used to enhance the feature response patterns.

[0136] The primary fusion feature and the shallow texture feature are fused through the feature selection gating mechanism in the decoder to obtain the target fusion feature. During the fusion process, the feature selection gating mechanism adaptively selects and enhances the feature channels related to spectral discrimination and the feature channels related to microtexture discrimination.

[0137] The target fusion features are respectively input into three parallel prediction heads of the lightweight image segmentation model. Through the three parallel prediction heads, the pixel masks of water bodies, floating garbage, and green algae in the image to be processed are output simultaneously.

[0138] Optionally, the identification module is further configured to:

[0139] Obtain the geographic location identifier of the target area and the seasonal identifier corresponding to the observation time window;

[0140] Based on the geographic location identifier and the seasonal identifier, the corresponding target spectral index set is obtained from the spectral index selection rule library preset on the satellite. The spectral indexes contained in the target spectral index set are used as the multispectral indexes of the target image area. The spectral index selection rule library defines the most effective lightweight spectral index combination for identifying black and odorous water bodies under different regional and seasonal conditions.

[0141] Calculate the spectral index corresponding to each spectral index based on the target image region;

[0142] All spectral indices corresponding to the target image region are input into a preset decision tree classification model with multi-level judgment nodes. The decision tree classification model makes judgments sequentially and outputs the black and odorous water pollution level of each pixel in the target image region. The judgment threshold of the decision tree classification model is pre-configured based on the water background characteristics and seasonal parameters of the target image region.

[0143] Based on the mask of the exposed water area, spatially connected water pixels are aggregated into independent candidate water patches;

[0144] For each candidate water body patch, the total area of ​​pixels inside the candidate water body patch that are determined to be black and odorous water bodies is counted.

[0145] If the total area ratio of any candidate water body patch is lower than a preset threshold, then the candidate water body patch is determined to be a normal water body patch, and pixels in the candidate water body patch that are determined to be black and odorous water bodies are removed. Otherwise, the candidate water body patch is determined to be a black and odorous water body patch. Based on the black and odorous water body patch, the black and odorous water body identification result is determined, and based on the black and odorous water body pollution level of each pixel in the candidate water body patch that is determined to be a black and odorous water body, the final pollution level of the black and odorous water body identification result is determined.

[0146] Optionally, the image filtering module is used for:

[0147] Based on the observation time window in the mission instructions, candidate images whose acquisition time falls entirely within the observation time window are selected from the image data stored on the satellite.

[0148] For each candidate image, calculate the spatial overlap between the geographic coverage of the candidate image and the target area in the task instruction;

[0149] Candidate images with spatial overlap greater than or equal to a preset overlap threshold are identified as spatiotemporally matched images to be processed.

[0150] Optionally, the processing parameters include task priority; the system further includes a configuration optimization module; the configuration optimization module is used to:

[0151] Before performing adaptive on-board processing on the image to be processed, the current overall resource load status is monitored in real time. The overall resource load status is determined based on at least one of the following: currently available memory, CPU utilization, remaining power budget, and the expected computational complexity of the current image to be processed.

[0152] The overall resource load status is compared with multiple preset resource level thresholds to determine the target resource level. Based on the target resource level and the task priority defined in the task instruction, the lightweight image segmentation model and the decision tree classification model are optimized and configured.

[0153] Wherein, when the target resource level indicates a mild resource shortage level and the task priority is normal, the first level of optimization is performed, and the first level of optimization is to suspend the calculation of the redundancy spectral index in the decision tree classification model;

[0154] When the target resource level indicates a moderate resource shortage level, or when the task priority is high, the first-level optimization and the second-level optimization are performed. The second-level optimization is to dynamically skip the preset non-critical feature extraction layer in the lightweight image segmentation model.

[0155] When the target resource level indicates that the current level is severe resource shortage, the first level optimization, the second level optimization, and the third level optimization are performed. The third level optimization is to determine the high-risk water body areas in the image to be processed based on the city's prior geographic information, so as to perform adaptive on-board processing only on the high-risk water body areas.

[0156] Optionally, the value assessment model includes a black and odorous water body scoring sub-model for scoring the black and odorous water body identification results; the black and odorous water body scoring sub-model is as follows:

[0157] S = α × G + β × log(A);

[0158] Where S represents the score of the black and odorous water body identification result, G represents the pollution level, A represents the area of ​​the black and odorous water body identification result, and α and β are configurable weight parameters.

[0159] It should be noted that other corresponding descriptions of the functional units involved in the real-time identification and information transmission system for black and odorous water bodies in satellite cities provided in this application embodiment can be found in the following references. Figures 1 to 3 The corresponding descriptions in the method will not be repeated here.

[0160] This application also provides a computer device, specifically a personal computer, server, network device, etc. This computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces. Figure 5 The diagram illustrates a computer device in which the processor, specifically a spaceborne AI processor, provides computational and control capabilities. The device's memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores location information. The device's network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements the steps in the various method embodiments.

[0161] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0162] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0163] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0165] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0167] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for real-time identification and information downlink of black and odorous water bodies in cities from space, characterized in that, include: Receive and parse standardized mission instructions from the ground, wherein the mission instructions define the target area, observation time window and processing parameters; Based on the target area and observation time window in the mission instructions, select spatiotemporally matching images to be processed from the image data stored on the satellite. According to the processing parameters in the task instruction, adaptive on-board processing is performed on the image to be processed to obtain a set of identification results, including black and odorous water body identification results and corresponding pollution levels, floating garbage identification results, and green algae and duckweed identification results; For each identification result in the identification result set, the identification result is scored using a value assessment model. The score for the black and odorous water body identification result is calculated based on the corresponding pollution level and area, while the scores for the floating garbage identification result and the green algae and duckweed identification result are calculated based on their respective corresponding areas. The scores of all identification results are sorted, high-value identification results are selected based on the sorting results, the high-value identification results are compressed, and they are prioritized for downlink transmission to the ground via the satellite-to-ground data transmission channel. The compression of the high-value identification results includes: The high-value identification results are categorized into black and odorous water body identification results, floating garbage identification results, and green algae and duckweed identification results. For the identification results of floating garbage and the identification results of green algae and duckweed, core attributes are encapsulated for each identification result, and a lightweight structured data file is generated based on the encapsulated core attributes. The core attributes include a unique identifier, spatial coordinates representing geographical location, and area of ​​the map patch. For the black and odorous water body identification results, a vector boundary data file describing the precise spatial range, a thematic map data file for visualization, and a statistical table data file recording detailed attributes are generated for each black and odorous water body patch in the black and odorous water body identification results. The value score of each black and odorous water body patch is calculated, and all black and odorous water body patches are divided into multiple importance levels based on the value score. For each black and odorous water body patch, based on the importance level corresponding to the black and odorous water body patch, the vector boundary data file, thematic map data file, and statistical table data file corresponding to the black and odorous water body patch are simplified and compressed.

2. The method according to claim 1, characterized in that, The adaptive on-board processing of the image to be processed yields a set of identification results, including black and odorous water body identification results and corresponding pollution levels, floating garbage identification results, and green algae and duckweed identification results. The preset lightweight image segmentation model is called to process the image to be processed, and the pixel masks of water bodies, floating garbage, and green algae and duckweed in the image to be processed are output simultaneously. From the water body pixel mask, remove the pixel areas that overlap with the garbage floating pixel mask and the green algae and duckweed pixel mask to obtain the exposed water body area mask; For the target image region corresponding to the mask of the exposed water body area, calculate the multispectral index of the target image region, and input the multispectral index into the pre-trained decision tree classification model for judgment to obtain the black and odorous water body identification result and the corresponding pollution level; The garbage floating identification result and the green algae and duckweed identification result are determined respectively based on the pixel mask of the floating garbage and the pixel mask of the green algae and duckweed; The identification results of the black and odorous water bodies, along with their corresponding pollution levels, the identification results of the floating garbage, and the identification results of the green algae and duckweed, are combined into the identification result set.

3. The method according to claim 2, characterized in that, The process involves calling a pre-set lightweight image segmentation model to process the image to be processed, and simultaneously outputting pixel masks for water bodies, floating garbage, and green algae / duckweed in the image to be processed, including: The image to be processed is input into the encoder of the lightweight image segmentation model, and the multi-scale features of the image to be processed are extracted and output simultaneously. The multi-scale features include shallow texture features, mid-level semantic features and deep abstract features. The multi-scale features are input into the decoder of the lightweight image segmentation model. Through the spatial and channel dual attention module in the decoder, the regions representing real water bodies and attachments in the deep abstract features are strengthened, while the regions representing typical interferences in the deep abstract features are suppressed. An attention weight map is generated based on the processing results. The attention weight map and the deep abstract features are multiplied element-wise by the product unit in the decoder to obtain spatially weighted deep guided features. The deep guiding features and the mid-level semantic features are fused together by the feature fusion unit in the decoder to obtain primary fused features. The mid-level semantic features include feature response patterns that are sensitive to the morphology of water body boundaries and spatial connectivity. The primary fused features are used to enhance the feature response patterns. The primary fusion feature and the shallow texture feature are fused through the feature selection gating mechanism in the decoder to obtain the target fusion feature. During the fusion process, the feature selection gating mechanism adaptively selects and enhances the feature channels related to spectral discrimination and the feature channels related to microtexture discrimination. The target fusion features are respectively input into three parallel prediction heads of the lightweight image segmentation model. Through the three parallel prediction heads, the pixel masks of water bodies, floating garbage, and green algae in the image to be processed are output simultaneously.

4. The method according to claim 2, characterized in that, For the target image region corresponding to the mask of the exposed water body area, the multispectral index of the target image region is calculated, and the multispectral index is input into a pre-trained decision tree classification model for judgment, to obtain the black and odorous water body identification result and the corresponding pollution level, including: Obtain the geographic location identifier of the target area and the seasonal identifier corresponding to the observation time window; Based on the geographic location identifier and the seasonal identifier, the corresponding target spectral index set is obtained from the spectral index selection rule library preset on the satellite. The spectral indexes contained in the target spectral index set are used as the multispectral indexes of the target image area. The spectral index selection rule library defines the most effective lightweight spectral index combination for identifying black and odorous water bodies under different regional and seasonal conditions. Calculate the spectral index corresponding to each spectral index based on the target image region; All spectral indices corresponding to the target image region are input into a preset decision tree classification model with multi-level judgment nodes. The decision tree classification model makes judgments sequentially and outputs the black and odorous water pollution level of each pixel in the target image region. The judgment threshold of the decision tree classification model is pre-configured based on the water background characteristics and seasonal parameters of the target image region. Based on the mask of the exposed water area, spatially connected water pixels are aggregated into independent candidate water patches; For each candidate water body patch, the total area of ​​pixels inside the candidate water body patch that are determined to be black and odorous water bodies is counted. If the total area ratio of any candidate water body patch is lower than a preset threshold, then the candidate water body patch is determined to be a normal water body patch, and pixels in the candidate water body patch that are determined to be black and odorous water bodies are removed. Otherwise, the candidate water body patch is determined to be a black and odorous water body patch. Based on the black and odorous water body patch, the black and odorous water body identification result is determined, and based on the black and odorous water body pollution level of each pixel in the candidate water body patch that is determined to be a black and odorous water body, the final pollution level of the black and odorous water body identification result is determined.

5. The method according to claim 1, characterized in that, The step of selecting spatiotemporally matching images to be processed from the image data stored on the satellite according to the target area and observation time window in the mission instructions includes: Based on the observation time window in the mission instructions, candidate images whose acquisition time falls entirely within the observation time window are selected from the image data stored on the satellite. For each candidate image, calculate the spatial overlap between the geographic coverage of the candidate image and the target area in the task instruction; Candidate images with spatial overlap greater than or equal to a preset overlap threshold are identified as spatiotemporally matched images to be processed.

6. The method according to claim 2, characterized in that, The processing parameters include task priority; before performing adaptive on-board processing on the image to be processed, the method further includes: Real-time monitoring of the current overall resource load status, wherein the overall resource load status is determined based on at least one of the following: currently available memory, CPU utilization, remaining power budget, and the estimated computational complexity of the current image to be processed; The overall resource load status is compared with multiple preset resource level thresholds to determine the target resource level. Based on the target resource level and the task priority defined in the task instruction, the lightweight image segmentation model and the decision tree classification model are optimized and configured. Wherein, when the target resource level indicates a mild resource shortage level and the task priority is normal, the first level of optimization is performed, and the first level of optimization is to suspend the calculation of the redundancy spectral index in the decision tree classification model; When the target resource level indicates a moderate resource shortage level, or when the task priority is high, the first-level optimization and the second-level optimization are performed. The second-level optimization is to dynamically skip the preset non-critical feature extraction layer in the lightweight image segmentation model. When the target resource level indicates that the current level is severe resource shortage, the first level optimization, the second level optimization, and the third level optimization are performed. The third level optimization is to determine the high-risk water body areas in the image to be processed based on the city's prior geographic information, so as to perform adaptive on-board processing only on the high-risk water body areas.

7. The method according to claim 1, characterized in that, The value assessment model includes a black and odorous water body scoring sub-model for scoring the identification results of the black and odorous water bodies; the black and odorous water body scoring sub-model is as follows: S = α × G + β × log(A); Where S represents the score of the black and odorous water body identification result, G represents the pollution level, A represents the area of ​​the black and odorous water body identification result, and α and β are configurable weight parameters.

8. A real-time identification and information transmission system for urban black and odorous water bodies on satellite, characterized in that, include: The instruction receiving module is used to receive and parse standardized mission instructions from the ground, wherein the mission instructions define the target area, observation time window and processing parameters; The image filtering module is used to filter out spatiotemporally matching images to be processed from the image data stored on the satellite according to the target area and observation time window in the mission instructions. The identification module is used to perform adaptive on-board processing on the image to be processed according to the processing parameters in the task instruction, and obtain an identification result set including the identification results of black and odorous water bodies and the corresponding pollution levels, garbage floating identification results, and green algae and duckweed identification results; The scoring calculation module is used to score each identification result in the identification result set using a value assessment model. The score of the black and odorous water body identification result is calculated based on the corresponding pollution level and area, and the scores of the floating garbage identification result and the green algae and duckweed identification result are calculated based on their respective areas. The downlink module is used to sort the scores of all recognition results, filter out high-value recognition results based on the sorting results, compress the high-value recognition results, and prioritize their downlink to the ground via the satellite-to-ground data transmission channel. The downlink module is also used for: The high-value identification results are categorized into black and odorous water body identification results, floating garbage identification results, and green algae and duckweed identification results. For the identification results of floating garbage and the identification results of green algae and duckweed, core attributes are encapsulated for each identification result, and a lightweight structured data file is generated based on the encapsulated core attributes. The core attributes include a unique identifier, spatial coordinates representing geographical location, and area of ​​the map patch. For the black and odorous water body identification results, a vector boundary data file describing the precise spatial range, a thematic map data file for visualization, and a statistical table data file recording detailed attributes are generated for each black and odorous water body patch in the black and odorous water body identification results. The value score of each black and odorous water body patch is calculated, and all black and odorous water body patches are divided into multiple importance levels based on the value score. For each black and odorous water body patch, based on the importance level corresponding to the black and odorous water body patch, the vector boundary data file, thematic map data file, and statistical table data file corresponding to the black and odorous water body patch are simplified and compressed.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images