A multi-source data fusion watershed water quality dynamic monitoring system and method

Abstract

The present application relates to the technical field of water quality monitoring, and particularly relates to a multi-source data fusion watershed water quality dynamic monitoring system and method. The method comprises the following steps: obtaining remote sensing images, water quality monitoring data, flow monitoring data and video monitoring data of a target watershed; constructing a water flow trajectory field based on the flow monitoring data and generating a pollution migration prediction path according to the water flow direction; extracting a continuous water body pixel sequence in the remote sensing images along the pollution migration prediction path, calculating the spectral gradient intensity between adjacent pixels, and generating a spectral weak disturbance sequence. The present application performs pixel sequence reconstruction and spectral gradient cumulative enhancement processing on multispectral remote sensing images based on water flow path constraints, focuses and continuously strengthens the expression of the dispersed and weakly changed water body pollution spectrum disturbance in space, so as to realize high-precision, continuous extraction and stable identification of the pollution plume front in the image domain.

Description

Technical Field

[0001] This invention relates to the field of water quality monitoring technology, and in particular to a watershed water quality dynamic monitoring system and method that integrates multi-source data. Background Technology

[0002] Watershed water quality monitoring typically relies on multi-source data, including remote sensing imagery, water quality sensors, flow monitoring stations, and video surveillance. Remote sensing imagery provides broad spatial coverage information, water quality monitoring data reflects changes in pollutant concentrations, flow data characterizes water transport conditions, and video data supplements local dynamic observation information. In practical applications, significant differences exist among these data sources in temporal and spatial resolution, as well as data representation formats. This makes it difficult to achieve unified spatiotemporal alignment and effective fusion of multi-source data, thus affecting the ability to continuously characterize pollution diffusion processes.

[0003] Furthermore, existing water quality analysis methods based on remote sensing images mostly focus on pixel-level spectral feature extraction, lacking the ability to structurally model pollution migration processes under the constraint of water flow direction. In complex river network environments, these methods are prone to problems such as discontinuous pollution boundary identification, difficulty in extracting weak pollution signals, and significant local noise interference. At the same time, single-data-source methods are difficult to dynamically correct and verify the pollution plume leading edge, resulting in insufficient stability of monitoring results during temporal evolution. Summary of the Invention

[0004] Therefore, it is necessary to provide a watershed water quality dynamic monitoring system and method that integrates multi-source data to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a method for dynamic monitoring of watershed water quality through multi-source data fusion is provided, the method comprising the following steps: Step S1: Acquire remote sensing images, water quality monitoring data, flow monitoring data, and video monitoring data of the target watershed; Step S2: Construct a water flow trajectory field based on flow monitoring data and generate a pollution migration prediction path according to the water flow direction; Step S3: Extract continuous water cell sequences from remote sensing images along the pollution migration prediction path, calculate the spectral gradient intensity between adjacent cells, and generate a spectral weak perturbation sequence. Step S4: Identify the marked area at the leading edge of the pollution plume based on the parameter variation trends of the weak spectral perturbation sequence and water quality monitoring data; Step S5: Dynamically verify the marked area at the leading edge of the contamination plume using video monitoring data, and correct the position of the leading edge boundary; Step S6: Generate watershed water quality dynamic monitoring results through the temporal migration process of the pollution plume leading edge marked area.

[0006] This invention also provides a multi-source data fusion-based watershed water quality dynamic monitoring system for executing the above-described multi-source data fusion-based watershed water quality dynamic monitoring method. The multi-source data fusion-based watershed water quality dynamic monitoring system includes: The data acquisition module is used to acquire remote sensing images, water quality monitoring data, flow monitoring data, and video monitoring data of the target watershed. The prediction module is used to construct a water flow trajectory field based on flow monitoring data and generate pollution migration prediction paths according to the direction of water flow. The analysis module is used to extract continuous water cell sequences from remote sensing images along the pollution migration prediction path, calculate the spectral gradient intensity between adjacent cells, and generate a spectral weak perturbation sequence. The labeling module is used to identify the labeled area at the leading edge of the pollution plume based on the parameter variation trend of the weak spectral perturbation sequence and water quality monitoring data; The verification module is used to dynamically verify the marked area at the leading edge of the contamination plume using video monitoring data and correct the position of the leading edge boundary. The display module is used to generate dynamic monitoring results of watershed water quality by observing the temporal migration process of the marked area at the leading edge of the pollution plume.

[0007] The beneficial effects of this invention lie in constructing an image sequence processing framework based on water flow path constraints by refining the processing of multi-source remote sensing images and extracting weakly changing pollution information. This framework achieves ordered reconstruction and continuous representation of water pixels in the flow direction by sampling remote sensing images along the river channel and unfolding them using multispectral bands. Furthermore, it calculates spectral gradient sequences based on the spectral response differences of adjacent pixels and aggregates and enhances the originally scattered, low-contrast spectral perturbation information in local space through gradient accumulation mapping and gradual compression enhancement, thereby improving the distinguishability of the pollution plume leading edge. Simultaneously, it introduces a spectral perturbation focusing mechanism to spatially compress and structurally reorganize continuously enhanced regions, enabling weakly changing pollution signals to form stable high-response regions in the image domain. Based on this, it eliminates isolated noise pixels through perturbation continuity and gradient preservation analysis, effectively preserving and optimizing the representation of the true pollution diffusion boundary. This significantly improves the extraction accuracy, boundary continuity, and anti-interference capability of the pollution plume leading edge in remote sensing images, achieving highly reliable image processing and dynamic recognition effects for watershed water quality changes. Attached Figure Description

[0008] Figure 1 A schematic diagram illustrating the steps of a multi-source data fusion method for dynamic monitoring of watershed water quality; Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3. Figure 3 This is a schematic diagram of the results of watershed water quality monitoring and analysis in one embodiment; Figure 4 This is a schematic diagram of a weak spectral perturbation extraction process according to one embodiment; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0009] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0010] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0011] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0012] To achieve the above objectives, please refer to Figures 1 to 4 A method for dynamic monitoring of watershed water quality through multi-source data fusion, the method comprising the following steps: Step S1: Acquire remote sensing images, water quality monitoring data, flow monitoring data, and video monitoring data of the target watershed; Step S2: Construct a water flow trajectory field based on flow monitoring data and generate a pollution migration prediction path according to the water flow direction; Step S3: Extract continuous water cell sequences from remote sensing images along the pollution migration prediction path, calculate the spectral gradient intensity between adjacent cells, and generate a spectral weak perturbation sequence. Step S4: Identify the marked area at the leading edge of the pollution plume based on the parameter variation trends of the weak spectral perturbation sequence and water quality monitoring data; Step S5: Dynamically verify the marked area at the leading edge of the contamination plume using video monitoring data, and correct the position of the leading edge boundary; Step S6: Generate watershed water quality dynamic monitoring results through the temporal migration process of the pollution plume leading edge marked area.

[0013] In one embodiment, taking a suspected industrial wastewater discharge incident in a downstream river basin of a city as an example, details can be found in [reference needed]. Figure 3 The monitoring area was the section of the river from the downstream industrial park to the lake inlet, acquiring multi-source data for the same time period. Remote sensing imagery combined Sentinel-2 multispectral imagery with low-altitude UAV-captured images to capture anomalies in river surface color and the distribution of suspended matter. Water quality monitoring data came from five automatic monitoring stations deployed along the river, including minute-level changes in COD, ammonia nitrogen, and turbidity. Flow monitoring data was provided by upstream hydrological stations, including cross-sectional flow velocity and water level changes. Video monitoring data came from fixed cameras on bridges and riverbanks to capture sudden color changes and mixing and diffusion phenomena in the local water body. All data were denoised and standardized after being aligned with unified timestamps to form a unified spatiotemporal reference dataset.

[0014] Using the upstream inlet as the starting reference point, a continuous hydrodynamic vector field is constructed by spatially interpolating the flow monitoring cross-sectional data. Further modifications to the vector field are made by incorporating river topographic constraints (such as bends, bifurcations, and dam structures), resulting in a constrained water flow trajectory field. Based on this, the area near the industrial park's discharge outlet is considered a potential pollution source. Path tracking is performed along the main flow direction, and high-probability diffusion paths are selected using a velocity-weighted factor. Finally, a pollution migration prediction path pointing towards the lake inlet is generated, reflecting the dominant diffusion channel of pollutants under current hydrodynamic conditions.

[0015] The predicted pollution migration path is mapped onto the remote sensing image coordinate system, and sampling is performed at fixed spatial intervals (e.g., 10m) to extract the corresponding water body pixel sequence. Within a range of approximately 2km downstream of the industrial park's discharge outlet, a spectral change in the water body, gradually changing from deep blue to grayish-green, can be observed. Multispectral reflectance differences are calculated for adjacent pixels, with a focus on analyzing changes in the green and near-infrared bands. When the spectral difference is below an adaptive threshold, it is identified as a weakly spectrally disturbed segment. This segment typically corresponds to the spatial location of the pollution plume transitioning from the core high-concentration area to the diffusion zone, thus forming a continuous sequence of weakly spectrally disturbed areas.

[0016] Using the changes in ammonia nitrogen concentration at different monitoring stations downstream of the industrial discharge outlet as a reference, a time-synchronous analysis was performed with a weak spectral perturbation sequence. For example, at monitoring station No. 3 downstream of the outlet, there was a period when the ammonia nitrogen concentration rapidly increased from 1.2 mg / L to 2.8 mg / L, and at the same time, the corresponding area in the remote sensing image showed enhanced spectral perturbation. Through sliding time window analysis, it was found that the changes in spectral perturbation lagged behind the changes in water quality concentration by about 10–20 minutes, which determined the transition zone of the pollution diffusion front. Further, combined with spatial continuity constraints, isolated anomalies were eliminated, and finally, a pollution plume leading edge marker area was identified, which is located in the river section about 1.5 km to 3 km downstream of the discharge outlet.

[0017] Frame-by-frame analysis of video data acquired by bridge cameras revealed a clear boundary line in the river section corresponding to the leading edge marker area of ​​the pollution plume, showing a gradual transition from clear to turbid water. Extraction of the water surface flow trajectory using optical flow revealed local backflow at river bends, causing a slight forward shift of the leading edge boundary in remote sensing identification. By overlaying and comparing the extracted true turbid boundary with the remotely sensed boundary, and applying weighted correction to the deviation area, the leading edge boundary was shifted back approximately 80–120 meters upstream, thereby improving boundary positioning accuracy.

[0018] By sequentially cascading the marker regions at the leading edge of the pollution plume acquired at different times, the migration trajectory of pollutants gradually advancing along the river from industrial discharge outlets to lake inlets can be obtained. By calculating the leading edge propagation velocity (e.g., 0.3–0.6 m / s) and the changes in diffusion width, a spatiotemporal evolution model of the pollution diffusion process can be reconstructed, and corresponding dynamic monitoring result maps can be generated, including pollution diffusion path maps, leading edge migration animation sequences, and water quality grade change distribution maps.

[0019] As an example of the present invention, reference is made to... Figure 2 As shown, step S3 in this example includes: Step S31: Based on the pollution migration prediction path, perform path constraint sampling on the remote sensing image, extract continuous water body pixels on the corresponding path, and construct a path pixel sequence according to the water flow direction; Step S32: Perform multispectral band expansion on the path pixel sequence to obtain the spectral response curve corresponding to each pixel; Step S33: Calculate the spectral variation amplitude between adjacent pixels based on the spectral response curve, and generate a spectral gradient sequence; Step S34: Perform gradient accumulation mapping on the spectral gradient sequence along the water flow direction to superimpose multiple continuously changing gradients to generate a gradient accumulation field; Step S35: Gradual compression enhancement is applied to the gradient accumulation field to compress the continuous and dispersed spectral changes into a local region, generating a spectral perturbation focusing map; Step S36: Extract the continuously enhanced regions in the spectral perturbation focusing map, and calculate the extension length, gradient preservation degree, and perturbation continuity of the continuously enhanced regions along the pollution migration prediction path to generate a spectral weak perturbation sequence.

[0020] In one embodiment, a pollution migration prediction path, starting from an industrial discharge outlet and ending at a lake inlet, is used as a reference. This path is projected onto a remote sensing image spatial coordinate system, and constrained sampling is performed along the path direction at fixed spatial intervals (e.g., 10m or one pixel scale). For each sampling point location, corresponding water body pixels are extracted, and shoreline, building shadows, and non-water body pixel interference are removed, thereby forming a continuous water body pixel sequence.

[0021] refer to Figure 4 For each water pixel obtained in step S31, multispectral features are deconstructed to extract its reflectance values ​​in the visible (blue, green, red), near-infrared, and short-wave infrared bands, and the corresponding spectral response curves are constructed. Taking the downstream area of ​​an industrial discharge outlet as an example, as pollutants diffuse, the reflectance in the green band gradually increases, while the near-infrared band shows a significant attenuation, thus forming a set of spectral response curves with continuous variation characteristics, which are used to characterize the spatial evolution trend of water pollution levels.

[0022] The spectral response curves of adjacent water body pixels are differentially calculated, and their variation amplitude is quantified using spectral angle difference or Euclidean distance. As the flow moves downstream from the discharge outlet, a "gradual change-abrupt change-gradual change" structural characteristic of spectral changes can be observed. A spectral gradient sequence is obtained through point-by-point calculation to characterize the strength distribution of spectral disturbances during pollution diffusion.

[0023] Using the direction of water flow as a constraint axis, the spectral gradient sequence is cumulatively weighted. For multiple consecutive gradient segments with similar trends, a sliding window method is used to superimpose them, forming a gradient accumulation effect, thereby enhancing the structural continuity of the pollution diffusion area. For example, in the 1km–3km range downstream of an industrial discharge outlet, multiple weakly changing gradients are superimposed to form a significant cumulative response zone, reflecting the main channel area of ​​pollutant diffusion.

[0024] Nonlinear compression enhancement processing is applied to the gradient accumulation field to enhance medium-to-high accumulation regions and suppress low-intensity background regions, thereby creating a spatial focusing effect on the originally dispersed weak spectral variations. Specifically, multiple low-amplitude variation regions are spatially redistributed to form high-contrast aggregation bands near the pollution migration path, thus generating a spectral perturbation focusing map.

[0025] Connectivity analysis was performed on the spectral perturbation focusing map to extract spatial regions with continuous enhancement characteristics, and projection measurements were performed along the predicted pollution migration path. The extension length characterizes the spatial span of the pollution impact range; gradient preservation measures the stability of spectral changes during propagation; and perturbation continuity describes the continuity of the enhanced region along the path direction. When all three indicators simultaneously meet preset threshold conditions, the region is defined as a weak spectral perturbation sequence, used to characterize the stable transition zone at the leading edge of the pollution plume.

[0026] In another embodiment, the preset threshold condition can be determined to be satisfied when the extension length is within the range of 0.6 to 0.9 times the pollution migration prediction path length, the gradient preservation degree is greater than 0.65 and less than or equal to 1, and the perturbation continuity is greater than or equal to 0.7.

[0027] Preferably, before performing path-constrained sampling of the remote sensing image based on the pollution migration prediction path, the method further includes: By using a multispectral imaging camera mounted on a drone or satellite platform to perform orbital flight or overpass scanning of the target watershed, and adjusting the pitch and heading angles of the multispectral imaging camera, the imaging optical axis is kept spatially aligned with the centerline of the river channel where the pollution migration prediction path is located. Set the exposure time and gain parameters of the CCD / CMOS photosensitive array in the multispectral imaging camera, reduce the probability of overexposure in the near-infrared band based on the reflectivity characteristics of the water body, and simultaneously start the band switching mechanism to complete the sequential acquisition of visible light and near-infrared bands. During the imaging process, the imaging time is time-marked using satellite ephemeris data or UAV GPS positioning data, and the shutter triggering sequence is controlled according to the time mark to make the remote sensing image acquisition window correspond to the water flow migration time window. The system receives raw imaging data from the image acquisition window and stores it in real time. Then, through the data interface, the raw imaging data is rearranged into an initial image sequence along the river channel according to the geographic coordinates to obtain remote sensing images of the target watershed.

[0028] In one embodiment, a multispectral imaging camera mounted on an unmanned aerial vehicle (UAV) platform or a satellite platform performs orbital flight or transit scanning of the target watershed. For example, the UAV performs low-altitude cruise scanning along the river channel centerline at a flight altitude of 120m and a speed of 8m / s, while the satellite platform performs periodic transit imaging of the same river section at an orbital altitude of approximately 500km. During the imaging process, the pitch angle (-5° to +15°) and heading angle (dynamically corrected within ±30°) of the multispectral imaging camera are adjusted in real time by an attitude control system to keep the deviation between the imaging optical axis and the river channel centerline corresponding to the pollution migration prediction path within ≤1 pixel scale, thereby ensuring the consistency between the image acquisition direction and the pollution diffusion direction.

[0029] Adaptive adjustment of multispectral imaging parameters is implemented. Specifically, the exposure time of the CCD / CMOS sensor array is dynamically adjusted between 2ms and 12ms. When the water reflectance is greater than 0.35, the exposure time is reduced to below 3ms to suppress saturation in high-reflectance areas; when the water reflectance is less than 0.15, the gain is increased to +6dB to +12dB to enhance detail information in low-reflectance areas. Simultaneously, a multispectral band switching mechanism is activated during acquisition, allowing visible light (B, G, R) and near-infrared (NIR) bands to alternately acquire images in a 0.5s cycle, following the sequence of "visible light → near-infrared → visible light," thus forming a time-aligned multi-band image sequence and reducing the probability of overexposure in a single band (controlled to ≤5% of pixels).

[0030] The imaging time is time-stamped using satellite ephemeris data or UAV GPS positioning data, with temporal resolution accuracy controlled within ±0.1s. The time stamps are then matched with the water flow velocity field along the predicted pollution migration path. For example, when the average downstream water flow velocity is 0.6m / s, the time stamp can control the spatial displacement error between two adjacent image frames to ≤0.06m. Furthermore, the shutter trigger frequency is controlled between 1Hz and 5Hz based on the time stamp, ensuring the synchronization error between the remote sensing image acquisition window and the water flow migration time window is controlled within ±2s.

[0031] The raw imaging data undergoes real-time storage and spatial rearrangement. Specifically, each frame of remote sensing image data is stored in GeoTIFF format, along with latitude and longitude coordinates and timestamp information. During data rearrangement, image frames are sorted according to river channel mileage based on the pollution migration prediction path, and the spatial resolution is uniformly resampled to 0.5 m / pixel. Through path-constrained rearrangement, the originally discretely acquired image data is reconstructed into an initial image sequence continuously distributed along the river channel, with a sequence length covering, for example, a river section of 5 km to 15 km.

[0032] Preferably, the exposure time and gain parameters of the CCD / CMOS photosensitive array in the multispectral imaging camera are set to reduce the probability of overexposure in the near-infrared band based on the reflectivity characteristics of the water body, specifically as follows: The near-infrared band filter and CCD / CMOS photosensitive array in the multispectral imaging camera are used to perform initial imaging sampling of the target water area and acquire near-infrared band pixels. Based on a preset threshold for water reflectance, the gray values ​​of near-infrared pixels are segmented and statistically analyzed to extract the response difference range between high-reflectance shoreline areas and low-reflectance water areas. Based on the saturation pixel ratio of the high-reflectivity shoreline region, the exposure time of the CCD / CMOS photosensitive array is adjusted by decreasing the value, and the gain amplification factor is reduced simultaneously. In response to the shutter triggering process of a multispectral imaging camera, the exposure sequence is controlled so that the exposure time of the near-infrared channel is shorter than that of the visible light channel.

[0033] In one embodiment, a near-infrared filter and a CCD / CMOS photosensitive array in a multispectral imaging camera are used to perform initial imaging sampling of the target water area to obtain near-infrared image data. Specifically, the initial sampling resolution is set to 0.5 m / pixel, and the sampling area covers a range of approximately 2 km downstream of the industrial discharge outlet. The water reflection signal is separated into single bands using a near-infrared filter (center wavelength approximately 850 nm, bandwidth ±30 nm) to obtain the corresponding pixel grayscale distribution.

[0034] Based on a preset threshold for water reflectance (e.g., 0.15–0.35), segmented statistical analysis was performed on the grayscale values ​​of near-infrared pixels. Specifically, pixels were divided into high-reflectance shoreline areas and low-reflectance water body areas. The grayscale values ​​of high-reflectance shoreline areas were typically greater than 180 (8-bit standard), while the grayscale values ​​of low-reflectance water body areas were typically less than 90. By statistically analyzing the proportion of saturated pixels in the high-reflectance areas, a risk of local overexposure was identified when the proportion of saturated pixels exceeded 10%, and the response difference range between the high-reflectance and low-reflectance areas was extracted.

[0035] The exposure time of the CCD / CMOS image sensor is gradually adjusted based on the saturation pixel ratio in the high-reflectivity shoreline region, while simultaneously reducing the gain amplification factor. For example, when the saturation ratio is between 10% and 20%, the exposure time is gradually reduced from the initial 10ms to 6ms; when the saturation ratio exceeds 20%, the exposure time is further reduced to 3ms, and the gain factor is reduced from +12dB to +6dB to suppress signal overflow in the high-reflectivity region, thereby reducing the probability of overexposure in the near-infrared band.

[0036] In response to the shutter triggering process of a multispectral imaging camera, the exposure sequence is synchronously controlled to ensure that the exposure time of the near-infrared channel is always shorter than that of the visible light channel. For example, when the exposure time of the visible light channel is set to 8ms to 12ms, the exposure time of the near-infrared channel is limited to 3ms to 6ms, and the timing control module ensures that the difference between the start time of the two exposures is controlled within 1ms, thereby avoiding brightness drift and time misalignment between different bands.

[0037] Preferably, the methods for obtaining high-reflectivity shoreline areas and low-reflectivity water body areas include: The land-water boundary of the remote sensing image of the target watershed is segmented, the edge pixels with continuous spatial distribution are extracted, and the candidate regions of the water body boundary are determined based on the gray-level difference of the neighborhood of the edge pixels. For candidate regions of water body boundaries, calculate the difference in reflectance of each pixel in the near-infrared band and the red band. Based on the reflectivity difference, continuous reflectance pixels in the near-infrared band are selected to obtain low-reflectance water body areas. Spectral response analysis is performed on neighboring pixels of non-water body boundary candidate regions to extract pixels with near-infrared reflectance higher than the preset reflectance threshold and spatially continuous distribution, which are then marked as high-reflectance shoreline regions.

[0038] In one embodiment, the remote sensing image of the target watershed is subjected to land-water boundary segmentation processing. Specifically, the remote sensing image is initially segmented based on NDWI (Normalized Difference Water Index) or an edge detection operator to extract the boundary contours between water bodies and non-water bodies, and further, a set of edge pixels continuously distributed along the boundary is extracted. Subsequently, a 3×3 or 5×5 neighborhood gray-level difference analysis is performed on the edge pixels. When the gray-level difference within the neighborhood is greater than a preset threshold (e.g., ΔI≥25), it is defined as a candidate region for the water body boundary.

[0039] For pixels within the candidate region of the water body boundary, the difference in reflectance between the near-infrared (NIR) and red bands is calculated, specifically expressed as ΔR = R_NIR - R_Red. When ΔR is lower than a preset threshold (e.g., ΔR ≤ 0.08) and the pixel grayscale is consistently below 90, the pixel is determined to have low reflectance characteristics. Based on spatial connectivity constraints (e.g., 4-neighborhood or 8-neighborhood connectivity rules), continuously distributed pixel regions are selected to obtain the low-reflectance water body region.

[0040] For neighboring pixels of non-water body boundary candidate regions, spectral response enhancement analysis is performed. Specifically, the reflectance of pixels in the near-infrared band is statistically analyzed. When the reflectance is higher than a preset threshold (e.g., R_NIR ≥ 0.18 or gray value ≥ 180), it is initially marked as a high-reflectance candidate pixel. Further, combined with spatial continuity constraints, regions that meet the criteria of continuous distribution length greater than 5 pixels or area greater than a certain pixel number threshold (e.g., ≥ 20 pixels) are retained, while isolated high-value noise points are removed, thus defining this type of region as a high-reflectance shoreline region.

[0041] Preferably, step S36 includes: The spectral perturbation focus map was directionally scanned along the pollution migration prediction path. The continuously enhanced region was divided into multiple continuous perturbation units according to the water flow direction, and the spectral change state corresponding to each perturbation unit was recorded. Based on the residual relationship of spectral changes between adjacent perturbation units, the degree of influence of the previous perturbation unit on the next perturbation unit is calculated, and the perturbation inertia correlation value is generated. Recursive analysis was performed on the perturbation inertia correlation values ​​of multiple continuous perturbation units to identify perturbation chains with continuous transmission characteristics and to eliminate single-point abrupt spectral anomaly regions. Based on the gradual attenuation trend of spectral change amplitude in the perturbation chain segment, a weak pollution perturbation attenuation trajectory is constructed to obtain the range of perturbation retention along the migration path; Based on the disturbance inertia correlation value, the continuity of the decay trajectory, and the consistency of the path direction, the disturbance inertia weight is generated. The perturbation inertia weight is used to screen continuous perturbation units to form a spectral weak perturbation sequence.

[0042] In one embodiment, the spectral perturbation focus map is directionally scanned along the pollution migration prediction path. Specifically, with the river flow direction as the main axis, the focus map is divided into continuous spatial units at fixed intervals (e.g., 10m corresponds to one pixel unit scale), and defined as a set of perturbation units. For each perturbation unit, its corresponding spectral change state parameters are recorded, including spectral gradient values, local enhancement intensity, and spatial connectivity indices, thereby constructing a basic perturbation state sequence.

[0043] Based on the residual relationship of spectral changes between adjacent perturbation units, the degree of influence of the previous perturbation unit on the subsequent perturbation unit is calculated, generating a perturbation inertial correlation value. Specifically, by comparing the spectral gradient difference ΔG and the directional consistency angle θ between the i-th unit and the (i+1)-th unit, when ΔG≤0.15 and θ≤15°, the spectral perturbation of the previous unit is considered to have a strong residual effect and is assigned a high inertial correlation weight (e.g., 0.7–0.95); otherwise, it is reduced to below 0.2, thereby quantifying the "inertial continuation effect" in the perturbation propagation process.

[0044] Recursive analysis is performed on the inertial correlation values ​​of multiple continuous disturbance units. A recursive propagation model is constructed using a sliding window (e.g., window length of 5–9 units) to identify disturbance chains with continuous propagation characteristics. When the inertial correlation value in a certain chain is consistently higher than a preset threshold (e.g., ≥0.6), it is determined to have stable propagation characteristics. At the same time, single-point abrupt spectral anomaly regions, such as isolated high-value points caused by local reflection, noise, or floating objects, are eliminated to avoid misjudging the pollution diffusion path.

[0045] Based on the gradual attenuation trend of spectral variation amplitude in the disturbance chain segment, a weak pollution disturbance attenuation trajectory is constructed. Specifically, the spectral gradient sequence within the chain segment is subjected to exponential or linear fitting. If its attenuation coefficient k is between 0.3 and 0.8, the chain segment is considered to conform to the natural diffusion attenuation law, and a model of the maintenance range of the disturbance along the water flow direction is established accordingly.

[0046] A comprehensive disturbance inertia weight is constructed based on the disturbance inertia correlation value, attenuation trajectory continuity, and path direction consistency. Specifically, weight coefficients (e.g., 0.4, 0.3, 0.3) are assigned to the three types of indicators, and the final disturbance inertia weight value is obtained through weighted fusion. When the weight value is higher than 0.65, the disturbance unit is considered to have stable propagation capability; if it is lower than this threshold, it is considered as a weakly correlated disturbance or noise interference.

[0047] In this embodiment, continuous perturbation units are screened using perturbation inertia weighting to form a spectral weak perturbation sequence. The final retained weak perturbation sequence exhibits a continuous spatial distribution along the river channel and a stable decreasing trend in time, thus effectively characterizing the extended transition zone of the pollution plume front.

[0048] In another embodiment, establishing the model for the maintenance range of disturbances along the water flow direction specifically involves: based on the continuous disturbance chain segments selected in step S36, extracting their spatial distribution sequence along the pollution migration prediction path, and discretizing them into equally spaced disturbance sampling units according to the water flow direction (e.g., dividing each space unit into 10m segments). For each disturbance unit, record its corresponding spectral gradient intensity Gi, disturbance inertial correlation value Ii, and position index i, thereby constructing a basic disturbance spatial sequence dataset.

[0049] The spatial sequence of disturbances is subjected to directional normalization, with the river flow direction defined as the positive reference axis, and a one-dimensional distance coordinate system established with the industrial discharge outlet location as the origin. Based on this coordinate system, the relative distance di of each disturbance unit is calculated, and the spectral gradient intensity Gi is mapped to the distance dimension to form an initial distribution curve of the disturbance intensity as a function of the flow direction.

[0050] A disturbance attenuation fitting mechanism is introduced to perform function fitting on the initial distribution curve. Specifically, an exponential decay model or a piecewise linear model is used to fit Gi, resulting in the disturbance attenuation function f(d) = G0·e^(-λd), where λ is the attenuation coefficient. Taking actual monitoring data of the XX River as an example, when λ is in the range of 0.12 to 0.25, the spectral weakening trend during the pollution diffusion process can be well fitted, thus reflecting the decreasing law of disturbance in the direction of water flow.

[0051] The attenuation function is weighted and corrected based on the disturbance inertia correlation value Ii. Specifically, when Ii ≥ 0.7, the attenuation coefficient λ of the corresponding disturbance element is reduced (e.g., reduced by 10% to 20%) to indicate that the disturbance in this region has strong propagation inertia; when Ii < 0.4, the attenuation coefficient correction magnitude is increased to reflect its rapid dissipation characteristics, thereby realizing the coupled modeling of disturbance propagation intensity and inertial characteristics.

[0052] The termination point of disturbance attenuation is determined based on a disturbance intensity threshold (e.g., Gi ≤ 0.05 or less than 20% of the initial intensity), and the spatial distance corresponding to this termination point is defined as the disturbance influence boundary. By performing boundary statistical analysis on all disturbance chain segments, a persistence range interval of the disturbance along the water flow direction is formed. This interval is used to characterize the effective continuous propagation distance of the pollution spectral disturbance in the river channel, and finally, a disturbance persistence range model is obtained.

[0053] Preferably, the degree of influence of the preceding perturbation unit on the following perturbation unit is calculated based on the residual relationship of spectral changes between adjacent perturbation units as follows: Obtain the peak positions of spectral changes of adjacent perturbation units and determine the centroid of change corresponding to each perturbation unit; Determine whether the centroid of the change in the subsequent perturbation unit is located on the downstream side of the peak value of the spectral change in the previous perturbation unit; When the center of gravity of the subsequent disturbance unit is misaligned with the peak value of the previous disturbance unit, the misalignment relationship is determined as a disturbance borrowing relationship. The extent to which the influence of the previous disturbance element on the subsequent disturbance element is determined by the number of borrowings and the borrowing distance in the continuous disturbance element.

[0054] In one embodiment, the peak position and centroid position of spectral changes of adjacent perturbation units are obtained. Specifically, local extremum detection is performed on the spectral gradient sequence within each perturbation unit to determine its maximum response point as the peak position Pi of the spectral change; at the same time, a weighted average of all spectral change values ​​within the perturbation unit is calculated to obtain the centroid position Ci, which is used to characterize the overall distribution center of the spectral change of the perturbation unit.

[0055] The direction of water flow is defined as the downstream reference axis, and a one-dimensional downstream coordinate system is established based on the pollution migration prediction path. For adjacent disturbance units Ui and Ui+1, it is determined whether the centroid of change Ci+1 of the latter disturbance unit Ui+1 is located downstream of the spectral change peak Pi of the former disturbance unit Ui. When there is a positive shift of Ci+1 relative to Pi in the downstream direction, the disturbance is considered to have a propagation and continuation trend; otherwise, the disturbance propagation is considered to have weakened or reverse interference characteristics.

[0056] When a spatial misalignment occurs between the centroid of the subsequent perturbation unit and the peak value of the spectral change of the preceding perturbation unit, this misalignment is defined as a perturbation borrowing relationship. Specifically, if the downstream distance offset Δd between Ci+1 and Pi is greater than a preset threshold (e.g., Δd ≥ 1 to 3 pixel scales), a perturbation borrowing phenomenon is determined to have occurred, indicating that the spectral influence of the preceding perturbation unit is spatially transmitted to the subsequent unit but is not fully aligned with its peak response position.

[0057] The persistence of the impact is quantified based on the number of borrowings and the borrowing distance in a continuous perturbation unit. Specifically, the number of borrowings is set as N_borrow, and the borrowing distance is set as d_borrow. The two are weighted and fused to form the impact persistence index E: E = α·N_borrow + β·Σd_borrow, where α and β are weighting coefficients, for example, 0.6 and 0.4 respectively. When the E value is high, it indicates that the previous perturbation unit has a strong spectral residual effect on the subsequent perturbation unit; when the E value is low, it indicates that the continuity of perturbation propagation is weak or that random fluctuation interference exists.

[0058] Preferably, determining whether the centroid of the change in the subsequent perturbation unit is located downstream of the peak value of the spectral change in the preceding perturbation unit includes: Extract the pixel position with the largest spectral change amplitude in the previous perturbation unit and determine it as the preceding peak point; The spatial distribution range of all perturbation pixels within the next perturbation unit is statistically analyzed, and the center of concentration of spectral change energy is determined to generate the center of gravity of change. A downstream projection axis is established based on the pollution migration prediction path, and the preceding peak point and the change centroid point are mapped onto the downstream projection axis. Determine whether the projection position of the changing centroid point on the downstream projection axis exceeds the projection position corresponding to the previous peak point; If the change centroid is located on the downstream projection side of the preceding peak point, it is determined that the subsequent perturbation unit has received the perturbation dominance released by the previous perturbation unit, and the change centroid is determined to be located on the downstream side of the spectral change peak of the previous perturbation unit.

[0059] In one embodiment, the pixel position with the largest spectral change amplitude in the preceding perturbation unit is extracted and determined as the preceding peak point. Specifically, the spectral change gradient of all pixels in the preceding perturbation unit Ui is scanned, and the pixel coordinate Pi with the largest value is selected as the spectral change peak position to characterize the dominant response position of the perturbation unit. In this embodiment, this peak typically corresponds to the strongest spectral response point of the pollution plume within the unit, such as a region of abrupt change in near-infrared reflectance caused by COD or suspended matter concentration.

[0060] The spatial distribution range of all perturbation pixels within the next perturbation unit is statistically analyzed, and the concentration center of spectral change energy is determined, generating the centroid of change. Specifically, all perturbation pixels in the next perturbation unit Ui+1 are weighted according to their spectral change amplitude to construct a spatially weighted centroid model, which is expressed as the centroid of change Ci+1. This model is used to characterize the spatial concentration trend of the overall spectral change of the perturbation unit, thereby avoiding interference of a single extreme value on the spatial structure.

[0061] A downstream projection axis is established based on the pollution migration prediction path. Specifically, the river centerline is parameterized to construct a one-dimensional downstream coordinate system with the water flow direction as the positive direction. The preceding peak point Pi and the changing centroid point Ci+1 are projected onto this downstream projection axis to obtain the corresponding one-dimensional position coordinates Pi′ and Ci+1′, so as to eliminate the influence of two-dimensional spatial offset on direction discrimination.

[0062] Determine whether the projection position of the change centroid on the downstream projection axis exceeds the projection position corresponding to the previous peak point. Specifically, when Ci+1′>Pi′, it indicates that the centroid of the spectral change of the subsequent disturbance unit is located downstream of the peak point of the previous disturbance unit, that is, it extends positively along the direction of water flow propagation; when Ci+1′≤Pi′, it indicates that the disturbance has not formed an effective downstream extension or there are backflow interference characteristics.

[0063] If the centroid of change is located downstream of the projection of the preceding peak, it is determined that the subsequent disturbance unit has received the disturbance dominance released by the preceding disturbance unit, and the centroid of change is determined to be located downstream of the peak value of the spectral change of the preceding disturbance unit. Through this determination mechanism, the "sequence relationship of dominant propagation direction" of pollution disturbances in the river channel can be identified.

[0064] Preferably, establishing a downstream projection axis based on the pollution migration prediction path, and mapping the preceding peak point and the change centroid point onto the downstream projection axis includes: Continuous path nodes are extracted along the predicted pollution migration path and formed into a path node chain according to the direction of water flow; Based on the path node chain, the connection between adjacent nodes is spliced ​​end to end to establish the downstream projection axis and generate the disturbance receiving line. Search along the disturbance connection line for the path node closest to the previous peak point, and determine the path node as the peak connection point; Search along the disturbance connection line for the path node closest to the changed centroid point, and determine the path node as the centroid connection point; Based on the order of the peak connection point and the centroid connection point in the disturbance connection line, the preceding peak point and the changing centroid point are mapped onto the downstream projection axis.

[0065] In one embodiment, continuous path nodes are extracted along the pollution migration prediction path, and a path node chain is formed according to the water flow direction. Specifically, the pollution migration prediction path is discretized at equal intervals, starting from the industrial discharge outlet and ending at the lake inlet, for example, setting one path node every 10m, resulting in a path node chain containing multiple sequentially arranged nodes. The nodes are connected by hydrodynamic direction vectors, thereby ensuring that the node chain strictly conforms to the dominant water flow direction in space.

[0066] Based on the path node chain, the lines connecting adjacent nodes are spliced ​​end-to-end to establish a downstream projection axis and generate a disturbance bearing line. Specifically, the path nodes are connected sequentially to form a continuous polyline structure, and the polyline is smoothed using spline interpolation or piecewise linear fitting to form a continuous downstream projection axis structure. This downstream projection axis also serves as the disturbance bearing baseline.

[0067] The path node closest to the preceding peak point is searched along the disturbance receiving line, and this path node is determined as the peak receiving point. Specifically, the Euclidean distance or projected distance between the preceding peak point Pi and all path nodes is calculated, and the node with the smallest distance is selected as the peak receiving point Pi. It is used to project peak information in two-dimensional space onto the downstream axis structure, thereby reducing the impact of spatial noise.

[0068] The nearest path node to the changed centroid is searched along the disturbance receiving line, and this path node is identified as the centroid receiving point. Specifically, the same nearest neighbor matching process is performed on the changed centroid Ci+1 to determine its corresponding centroid receiving point Ci+1. In this way, the information of the disturbance centroid can be stably mapped to a path node chain consistent with the river channel structure.

[0069] Based on the order of the peak and centroid connection points in the disturbance connection line, the preceding peak points and the changing centroid points are mapped onto the downstream projection axis. Specifically, using the sequential index of the path node chain as a one-dimensional coordinate reference, when the centroid connection point is downstream of the peak connection point, the changing centroid point is mapped to the positive extension position of the downstream axis; otherwise, it is mapped to its upstream direction, thus completing the unified expression from two-dimensional spatial feature points to a one-dimensional downstream projection axis.

[0070] Of particular importance, mapping the preceding peak point and the changed centroid point onto the downstream projection axis according to the order of the peak point and the centroid point in the disturbance connection line also includes: Extract the continuous path segment between the peak bearing point and the center bearing point along the disturbance bearing line, and determine this path segment as the candidate yielding area; Analyze the distribution of spectral change amplitudes for each path node within the candidate yielding zone to determine the locations where the spectral change changes from enhancement to weakening. The location where the spectral change changes from enhancement to weakening is identified as the perturbation site; Centered on the disturbance yield point, a preset range is extended in both downstream and upstream directions to generate a disturbance yield zone; Dominance migration data is generated based on the distribution of the disturbance yield zone in the disturbance bearing line.

[0071] In one embodiment, a continuous path segment between the peak receiving point and the centroid receiving point is extracted along the disturbance receiving line, and this path segment is identified as a candidate yielding zone. Specifically, using the peak receiving point as the starting node and the centroid receiving point as the ending node, all path nodes between the two points are traversed node by node along the disturbance receiving line. For example, between the preceding peak receiving point 2.3 km downstream of the river industrial discharge outlet and the centroid receiving point 2.8 km downstream, there are a total of 52 continuous path nodes. The path segment consisting of these 52 nodes is identified as a candidate yielding zone. Since this area is located in the potential range where the dominant position of disturbance propagation changes, it is considered a key area for subsequent dominance transfer analysis.

[0072] The distribution of spectral variation amplitudes for each path node within the candidate yielding zone is analyzed to determine the location where the spectral variation transitions from enhancement to reduction. Specifically, the spectral gradient values ​​for each path node within the candidate yielding zone are extracted and a gradient evolution sequence is formed according to the downstream direction. For example, in a certain candidate yielding zone, the spectral variation amplitudes of consecutive nodes are 0.18, 0.24, 0.31, 0.37, 0.42, 0.39, 0.33, 0.27, and 0.21, respectively. It can be identified that the spectral variation amplitude reaches a local maximum at 0.42 and then begins to decrease continuously; therefore, the corresponding node is identified as the enhancement-reduction transition node.

[0073] The location where the spectral change transitions from enhancement to reduction is identified as a perturbation yield point. Specifically, when a path node simultaneously satisfies the condition that the spectral change amplitude of the preceding node continuously increases while the spectral change amplitude of the subsequent node continuously decreases, that node is marked as a perturbation yield point. For example, the local maximum node mentioned above corresponds to a river channel distance of 2.56 km, so this location is identified as a perturbation yield point. This location indicates that the dominant influence of the preceding perturbation unit reaches its peak and begins to decay, while the subsequent perturbation unit gradually takes over the dominance of perturbation propagation.

[0074] Centered on the disturbance yielding point, a predetermined range is extended in both the downstream and upstream directions to generate a disturbance yielding zone. Specifically, the node where the disturbance yielding point is located is taken as the central node, and a predetermined distance is extended upstream and downstream, respectively. For example, extending 30m upstream and 50m downstream forms a disturbance yielding zone with a total length of 80m. Furthermore, the extension can also be based on the number of path nodes, such as extending forward by 5 path nodes and backward by 8 path nodes, thus obtaining a disturbance yielding zone containing multiple consecutive nodes. This region is used to characterize the spatial buffer zone where disturbance dominance shifts, rather than a single spatial point.

[0075] Dominance transfer data is generated based on the distribution of the disturbance yielding region along the disturbance receiving line. Specifically, the positional ratio of the disturbance yielding region relative to the peak receiving point and the centroid receiving point is calculated. This is combined with the length of the yielding region, the rate of spectral decay within the yielding region, and the magnitude of change in the disturbance inertial correlation value before and after the yielding region to construct a dominance transfer index. For example, when the yielding region is located at 65% of the distance between the peak receiving point and the centroid receiving point, and the disturbance inertial correlation value decreases from 0.83 to 0.42, it can be determined that there is a significant dominance transfer in this region, and corresponding dominance transfer data is generated.

[0076] Dominance transfer data can include parameters such as the starting location of the migration, the location of the completed migration, the migration duration, the migration intensity level, and the stability of the migration direction. Among them, the migration duration reflects the spatial span of the disturbance dominance handover process; the migration intensity level reflects the significance of the dominance transfer; and the stability of the migration direction measures whether the dominance transfer continues along the pollution migration prediction path.

[0077] Most importantly, the predetermined range is extended from the disturbance relocation site in both the downstream and upstream directions as follows: Spectral crossover scanning is performed with the perturbation receptive site as the center, and the path nodes where the dominant feature of the preceding peak persists are searched along the perturbation receptive line in the countercurrent direction. The point at which the dominant feature of the preceding peak changes from continuous to discontinuous is defined as the peak exit point; Spectral crossover scanning is performed with the perturbation receptive site as the center, and the path nodes where the dominant feature of the changing center of gravity persists are searched along the perturbation receptive line in the downstream direction. The location where the dominant feature of the changing center of gravity changes from discontinuous to continuous is determined as the center of gravity access point; A disturbance yield zone is generated based on the path segment between the peak exit point and the center of gravity access point.

[0078] In one embodiment, a spectral crossover scan is performed centered on the perturbation yield point, searching for path nodes along the perturbation receiving line in the countercurrent direction where the preceding peak dominance characteristic persists. Specifically, the path node corresponding to the perturbation yield point is used as the starting point for the scan, and the spectral change state of each path node is analyzed back along the countercurrent direction. In this embodiment, the preceding peak dominance characteristic includes the peak influence ratio, peak inheritance intensity, and peak direction retention rate. The peak influence ratio characterizes the degree to which the current node is affected by the preceding peak perturbation; the peak inheritance intensity characterizes the ability of the current node to retain the spectral morphology of the preceding peak; and the peak direction retention rate characterizes the consistency between the perturbation propagation direction and the preceding peak propagation direction. For example, in the downstream area of ​​a river industrial discharge outlet, if the peak influence ratio of consecutive path nodes is higher than 65%, the preceding peak dominance characteristic is considered to persist.

[0079] As the reverse-flow scanning process proceeds, the rate of change of peak inheritance between adjacent path nodes is calculated. When multiple consecutive nodes maintain peak dominance characteristics, it indicates that the region is still in the preceding peak control stage; when the peak influence ratio of a certain path node drops below a preset threshold (e.g., below 40%), or the peak direction retention rate drops below 50%, it is considered that the preceding peak dominance characteristics have broken.

[0080] The point where the dominant characteristic of the preceding peak changes from continuous to discontinuous is defined as the peak exit point. For example, if the peak inheritance strength remains above 0.7 until the 8th path node upstream of the perturbation yield point, but drops to 0.35 at the 9th path node, then the boundary between the 8th and 9th path nodes is defined as the peak exit point. This location indicates that the preceding perturbation unit begins to relinquish its control over the subsequent propagation process.

[0081] A spectral crossover scan is performed centered on the perturbation yielding point, searching for path nodes along the perturbation receiving line in the downstream direction where the dominant characteristic of the changing center of gravity persists. Specifically, starting from the perturbation yielding point, the contribution of the changing center of gravity, the stability of the center of gravity clustering, and the control rate of the center of gravity diffusion are analyzed node by node in the downstream direction. Among them, the contribution of the changing center of gravity is used to characterize the degree to which the current node is dominated by the changing center of gravity; the stability of the center of gravity clustering is used to characterize the spatial concentration stability of the perturbation center of gravity; and the control rate of the center of gravity diffusion is used to characterize the ability of the changing center of gravity to guide the diffusion direction of subsequent perturbations.

[0082] In this embodiment, when the contribution of the changing centroid in the continuous path nodes is consistently higher than 60% and the centroid aggregation stability remains above 0.65, it is determined that the dominant feature of the changing centroid is continuously present; when the original discrete state gradually changes to a continuous stable state during the scanning process, it indicates that the disturbance propagation control center begins to shift from the preceding peak value to the changing centroid.

[0083] In this embodiment, the location where the dominant feature of the changing center of gravity transitions from discontinuous to continuous is defined as the center of gravity access point. For example, in the region downstream of the disturbance yield point, the center of gravity contributions of the first three nodes are 35%, 42%, and 48%, respectively, failing to form continuous control; while the fourth node begins to continuously reach 68%, 71%, and 75%, the fourth node is then defined as the center of gravity access point. This location indicates that the changing center of gravity begins to acquire stable disturbance dominance.

[0084] A disturbance yield zone is generated based on the path segment between the peak exit point and the center of gravity access point. Specifically, the peak exit point is used as the boundary for releasing disturbance dominance, and the center of gravity access point is used as the boundary for receiving disturbance dominance. All continuous path nodes between the two are extracted to form the disturbance yield zone. For example, when the peak exit point is located at a river channel length of 2.42 km and the center of gravity access point is located at a river channel length of 2.58 km, the path segment between 2.42 km and 2.58 km is determined as the disturbance yield zone.

[0085] The disturbance yield zone is not simply a spatial transition area, but a dynamic transition zone characterizing the shift in disturbance dominance from a preceding peak control mode to a changing center of gravity control mode. Within this zone, the influence of the preceding peak continuously diminishes, while the control capability of the changing center of gravity continuously strengthens, and the two together constitute the dominance transition zone in the propagation process of the pollution plume.

[0086] This invention also provides a multi-source data fusion-based watershed water quality dynamic monitoring system for executing the aforementioned multi-source data fusion-based watershed water quality dynamic monitoring method. The multi-source data fusion-based watershed water quality dynamic monitoring system includes: The data acquisition module is used to acquire remote sensing images, water quality monitoring data, flow monitoring data, and video monitoring data of the target watershed. The prediction module is used to construct a water flow trajectory field based on flow monitoring data and generate pollution migration prediction paths according to the direction of water flow. The analysis module is used to extract continuous water cell sequences from remote sensing images along the pollution migration prediction path, calculate the spectral gradient intensity between adjacent cells, and generate a spectral weak perturbation sequence. The labeling module is used to identify the labeled area at the leading edge of the pollution plume based on the parameter variation trend of the weak spectral perturbation sequence and water quality monitoring data; The verification module is used to dynamically verify the marked area at the leading edge of the contamination plume using video monitoring data and correct the position of the leading edge boundary. The display module is used to generate dynamic monitoring results of watershed water quality by observing the temporal migration process of the marked area at the leading edge of the pollution plume.

[0087] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0088] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for dynamic monitoring of watershed water quality through multi-source data fusion, characterized in that, Includes the following steps: Step S1: Acquire remote sensing images, water quality monitoring data, flow monitoring data, and video monitoring data of the target watershed; Step S2: Construct a water flow trajectory field based on flow monitoring data and generate a pollution migration prediction path according to the water flow direction; Step S3: Extract continuous water cell sequences from remote sensing images along the pollution migration prediction path, calculate the spectral gradient intensity between adjacent cells, and generate a spectral weak perturbation sequence. Step S4: Identify the marked area at the leading edge of the pollution plume based on the parameter variation trends of the weak spectral perturbation sequence and water quality monitoring data; Step S5: Dynamically verify the marked area at the leading edge of the contamination plume using video monitoring data, and correct the position of the leading edge boundary; Step S6: Generate watershed water quality dynamic monitoring results through the temporal migration process of the pollution plume leading edge marked area.

2. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Based on the pollution migration prediction path, perform path constraint sampling on the remote sensing image, extract continuous water body pixels on the corresponding path, and construct a path pixel sequence according to the water flow direction; Step S32: Perform multispectral band expansion on the path pixel sequence to obtain the spectral response curve corresponding to each pixel; Step S33: Calculate the spectral variation amplitude between adjacent pixels based on the spectral response curve, and generate a spectral gradient sequence; Step S34: Perform gradient accumulation mapping on the spectral gradient sequence along the water flow direction to superimpose multiple continuously changing gradients to generate a gradient accumulation field; Step S35: Gradual compression enhancement is applied to the gradient accumulation field to compress the continuous and dispersed spectral changes into a local region, generating a spectral perturbation focusing map; Step S36: Extract the continuously enhanced regions in the spectral perturbation focusing map, and calculate the extension length, gradient preservation degree, and perturbation continuity of the continuously enhanced regions along the pollution migration prediction path to generate a spectral weak perturbation sequence.

3. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 2, characterized in that, Before performing path-constrained sampling of remote sensing images based on pollution migration prediction paths, the following steps are also included: By using a multispectral imaging camera mounted on a drone or satellite platform to perform orbital flight or overpass scanning of the target watershed, and adjusting the pitch and heading angles of the multispectral imaging camera, the imaging optical axis is kept spatially aligned with the centerline of the river channel where the pollution migration prediction path is located. Set the exposure time and gain parameters of the CCD / CMOS photosensitive array in the multispectral imaging camera, reduce the probability of overexposure in the near-infrared band based on the reflectivity characteristics of the water body, and simultaneously start the band switching mechanism to complete the sequential acquisition of visible light and near-infrared bands. During the imaging process, the imaging time is time-marked using satellite ephemeris data or UAV GPS positioning data, and the shutter triggering sequence is controlled according to the time mark to make the remote sensing image acquisition window correspond to the water flow migration time window. The system receives raw imaging data from the image acquisition window and stores it in real time. Then, through the data interface, the raw imaging data is rearranged into an initial image sequence along the river channel according to the geographic coordinates to obtain remote sensing images of the target watershed.

4. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 3, characterized in that, To reduce the probability of overexposure in the near-infrared band by setting the exposure time and gain parameters of the CCD / CMOS photosensitive array in the multispectral imaging camera based on the reflectivity characteristics of water, the following steps are taken: The near-infrared band filter and CCD / CMOS photosensitive array in the multispectral imaging camera are used to perform initial imaging sampling of the target water area and acquire near-infrared band pixels. Based on a preset threshold for water reflectance, the gray values ​​of near-infrared pixels are segmented and statistically analyzed to extract the response difference range between high-reflectance shoreline areas and low-reflectance water areas. Based on the saturation pixel ratio of the high-reflectivity shoreline region, the exposure time of the CCD / CMOS photosensitive array is adjusted by decreasing the value, and the gain amplification factor is reduced simultaneously. In response to the shutter triggering process of a multispectral imaging camera, the exposure sequence is controlled so that the exposure time of the near-infrared channel is shorter than that of the visible light channel.

5. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 4, characterized in that, Methods for obtaining high-reflectivity shoreline areas and low-reflectivity water body areas include: The land-water boundary of the remote sensing image of the target watershed is segmented, the edge pixels with continuous spatial distribution are extracted, and the candidate regions of the water body boundary are determined based on the gray-level difference of the neighborhood of the edge pixels. For candidate regions of water body boundaries, calculate the difference in reflectance of each pixel in the near-infrared band and the red band. Based on the reflectivity difference, continuous reflectance pixels in the near-infrared band are selected to obtain low-reflectance water body areas. Spectral response analysis is performed on neighboring pixels of non-water body boundary candidate regions to extract pixels with near-infrared reflectance higher than the preset reflectance threshold and spatially continuous distribution, which are then marked as high-reflectance shoreline regions.

6. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 2, characterized in that, Step S36 includes: The spectral perturbation focus map was directionally scanned along the pollution migration prediction path. The continuously enhanced region was divided into multiple continuous perturbation units according to the water flow direction, and the spectral change state corresponding to each perturbation unit was recorded. Based on the residual relationship of spectral changes between adjacent perturbation units, the degree of influence of the previous perturbation unit on the next perturbation unit is calculated, and the perturbation inertia correlation value is generated. Recursive analysis was performed on the perturbation inertia correlation values ​​of multiple continuous perturbation units to identify perturbation chains with continuous transmission characteristics and to eliminate single-point abrupt spectral anomaly regions. Based on the gradual attenuation trend of spectral change amplitude in the perturbation chain segment, a weak pollution perturbation attenuation trajectory is constructed to obtain the range of perturbation retention along the migration path; Based on the disturbance inertia correlation value, the continuity of the decay trajectory, and the consistency of the path direction, the disturbance inertia weight is generated. The perturbation inertia weight is used to screen continuous perturbation units to form a spectral weak perturbation sequence.

7. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 6, characterized in that, Based on the residual relationship of spectral changes between adjacent perturbation units, the degree of persistence of the influence of the previous perturbation unit on the subsequent perturbation unit is calculated as follows: Obtain the peak positions of spectral changes of adjacent perturbation units and determine the centroid of change corresponding to each perturbation unit; Determine whether the centroid of the change in the subsequent perturbation unit is located on the downstream side of the peak value of the spectral change in the previous perturbation unit; When the center of gravity of the subsequent disturbance unit is misaligned with the peak value of the previous disturbance unit, the misalignment relationship is determined as a disturbance borrowing relationship. The extent to which the influence of the previous disturbance element on the subsequent disturbance element is determined by the number of borrowings and the borrowing distance in the continuous disturbance element.

8. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 7, characterized in that, Determining whether the centroid of the change in the subsequent perturbation unit is located downstream of the peak value of the spectral change in the preceding perturbation unit includes: Extract the pixel position with the largest spectral change amplitude in the previous perturbation unit and determine it as the preceding peak point; The spatial distribution range of all perturbation pixels within the next perturbation unit is statistically analyzed, and the center of concentration of spectral change energy is determined to generate the center of gravity of change. A downstream projection axis is established based on the pollution migration prediction path, and the preceding peak point and the change centroid point are mapped onto the downstream projection axis. Determine whether the projection position of the changing centroid point on the downstream projection axis exceeds the projection position corresponding to the previous peak point; If the change centroid is located on the downstream projection side of the preceding peak point, it is determined that the subsequent perturbation unit has received the perturbation dominance released by the previous perturbation unit, and the change centroid is determined to be located on the downstream side of the spectral change peak of the previous perturbation unit.

9. The method for dynamic monitoring of watershed water quality based on multi-source data fusion according to claim 7, characterized in that, Based on the pollution migration prediction path, a downstream projection axis is established, and the preceding peak points and the change centroid points are mapped onto the downstream projection axis, including: Continuous path nodes are extracted along the predicted pollution migration path and formed into a path node chain according to the direction of water flow; Based on the path node chain, the connection between adjacent nodes is spliced ​​end to end to establish the downstream projection axis and generate the disturbance receiving line. Search along the disturbance connection line for the path node closest to the previous peak point, and determine the path node as the peak connection point; Search along the disturbance connection line for the path node closest to the changed centroid point, and determine the path node as the centroid connection point; Based on the order of the peak connection point and the centroid connection point in the disturbance connection line, the preceding peak point and the changing centroid point are mapped onto the downstream projection axis.

10. A watershed water quality dynamic monitoring system based on multi-source data fusion, characterized in that, The watershed water quality dynamic monitoring system for performing the multi-source data fusion method as described in claim 1 includes: The data acquisition module is used to acquire remote sensing images, water quality monitoring data, flow monitoring data, and video monitoring data of the target watershed. The prediction module is used to construct a water flow trajectory field based on flow monitoring data and generate pollution migration prediction paths according to the direction of water flow. The analysis module is used to extract continuous water cell sequences from remote sensing images along the pollution migration prediction path, calculate the spectral gradient intensity between adjacent cells, and generate a spectral weak perturbation sequence. The labeling module is used to identify the labeled area at the leading edge of the pollution plume based on the parameter variation trend of the weak spectral perturbation sequence and water quality monitoring data; The verification module is used to dynamically verify the marked area at the leading edge of the contamination plume using video monitoring data and correct the position of the leading edge boundary. The display module is used to generate dynamic monitoring results of watershed water quality by observing the temporal migration process of the marked area at the leading edge of the pollution plume.

Images