Black and odorous water body remote sensing comprehensive research and judgment method, device and equipment and storage medium
By constructing a comprehensive evaluation model of black and odorous water bodies with multi-source data, combining information about black and odorous water bodies, eutrophication and potential pollution sources, the problem of incomplete identification in the existing technology is solved, and a more efficient and scientific black and odorous water body is achieved. Water body identification.
Patent Information
- Application Number
- CN202410961098.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-15
AI Technical Summary
In the prior art, the identification evaluation index of black and odorous water bodies is single, and the identification is not comprehensive and accurate enough. At the same time, the analysis and judgment of potential risk factors is lacking, resulting in insufficient identification accuracy and efficiency.
Using a comprehensive analysis and judgment method of multi-source data, combined with the black and odorous water body recognition model, eutrophied water body recognition model, remote sensing water quality parameter inversion model and potential pollution source risk extraction model, a comprehensive evaluation model of black and odorous water body was constructed through different indices and models. Combined with the identification of black and odorous water bodies, the entropy value method is used to determine the weight of each evaluation index.
It significantly improves the universality and accuracy of black and odorous water bodies recognition, takes into account the impact of eutrophication and potential pollution sources, improves the accuracy and portability of the identification, and ensures the objectivity and scientificity of the evaluation results.
Smart Images

Figure CN118966751A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of remote sensing technology and water environment monitoring technology, and in particular to a method, device, equipment and storage medium for comprehensive remote sensing analysis of black and odorous water bodies using multi-source data. Background Art
[0002] The current remote sensing identification technology for black and odorous water bodies mainly focuses on using remote sensing indicators such as the unique spectral characteristics and water color characteristics of black and odorous water bodies for accurate identification, and has developed a variety of indexes for the identification of black and odorous water bodies, such as NDBWI, DBWI, and BOI. These indices calculate and process the information of specific bands in remote sensing images to extract characteristic information related to black and odorous water bodies, thereby realizing the identification and monitoring of black and odorous water bodies. However, the causes of black and odorous water bodies are complex, and a single index cannot fully reflect their characteristics and changes. Therefore, it is necessary to adopt a comprehensive evaluation method to comprehensively analyze and integrate the information of multiple indices. By combining the advantages of different indices, black and odorous water bodies can be identified and monitored more comprehensively and accurately.
[0003] In addition, the formation and distribution of black and odorous water bodies have obvious geographical characteristics. In residential areas with high population density and many emission sources, as well as construction sites and broken rivers, these areas are more likely to become breeding grounds for black and odorous water bodies due to poor water flow and frequent pollution emissions. Current black and odorous water body identification algorithms often lack research and judgment on these potential risk factors. By increasing the analysis and evaluation of potential risk factors, the formation conditions and distribution of black and odorous water bodies can be judged more accurately, the portability of the algorithm can be increased, and the accuracy and efficiency of black and odorous water body identification can be improved. Summary of the invention
[0004] The main purpose of the present invention is to solve the problems in the prior art that the black and odorous water body identification and evaluation index is single, the black and odorous water body identification is not comprehensive and accurate, and at the same time, there is a lack of research and judgment on potential risk factors.
[0005] The first aspect of the present invention provides a remote sensing comprehensive analysis method for black and odorous water bodies, comprising:
[0006] Based on the black and odorous water body identification model, eutrophication water body identification model, remote sensing water quality parameter inversion model and potential pollution source risk extraction model, black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction were carried out respectively;
[0007] The black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction are used as evaluation indicators, the weight of each evaluation indicator is determined, and a comprehensive evaluation model for black and odorous water bodies is constructed;
[0008] The black and odorous water bodies are identified using the constructed comprehensive evaluation model of black and odorous water bodies;
[0009] The black and odorous water body identification model realizes the black and odorous water body identification by combining different black and odorous index models with on-site measured spectral information;
[0010] The eutrophication water body identification model is based on lake bloom distribution and water eutrophication degree to achieve eutrophication water body identification;
[0011] The remote sensing water quality parameter inversion model uses the prediction results of different types of base learners as the input of the meta-learner to achieve water quality prediction;
[0012] The potential pollution source risk extraction model adopts a multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value, and uses the kernel density analysis method to obtain the density area map and change trend of the potential pollution source risk points.
[0013] Optionally, in a first implementation of the first aspect of the present invention, the black and odorous water body identification model includes:
[0014] The black and odorous water bodies are initially extracted based on the black and odorous water index empirical threshold, including the black and odorous water body difference index model, the normalized difference black and odorous water body index model and the black and odorous water body index model, combined with historical spectral information;
[0015] The black and odorous water body difference index model includes:
[0016] DBWI=R rs (green)-R rs (blue)
[0017] The normalized difference black and odorous water index model includes:
[0018]
[0019] The black and odorous water index model includes:
[0020]
[0021] Among them, R rs Represents the reflectivity value of the corresponding band after atmospheric correction; R rs (green) represents the reflectance value of the green band after atmospheric correction; R rs (blue) indicates the reflectance value of the blue band after atmospheric correction; R rs (red) indicates the reflectance value of the red band after atmospheric correction;
[0022] Based on the preliminary extraction of black and odorous water bodies and combined with field measured data, the characteristic threshold of the black and odorous index is updated to accurately extract black and odorous water bodies.
[0023] Optionally, in a second implementation of the first aspect of the present invention, the eutrophic water body identification model includes:
[0024] Using the Normalized Difference Vegetation Index (NDVI) to identify the distribution of algal blooms in lakes:
[0025]
[0026] Among them, R NIR Represents the spectral reflectance value in the near-infrared band; R RED Indicates the spectral reflectance value of the red band;
[0027] The comprehensive nutrient status index TLI (∑) is used to evaluate the degree of eutrophication of water bodies:
[0028]
[0029] Wherein, TLI(j) represents the nutritional status parameter of the jth parameter; W ij represents the relevant weight of the nutritional status index of the jth parameter; r ij represents the correlation coefficient between the jth parameter and the benchmark parameter total phosphorus; m represents the number of evaluation parameters.
[0030] Optionally, in a third implementation of the first aspect of the present invention, the remote sensing water quality parameter inversion model includes:
[0031] Use the water quality spectrum library to train the remote sensing water quality parameter inversion model, and use the trained remote sensing water quality parameter inversion model to predict water quality;
[0032] The method of using a water quality spectrum library to train a remote sensing water quality parameter inversion model comprises:
[0033] Based on the water quality spectral library, k training sets of equal size were constructed using the k-fold cross-validation method;
[0034] When each base learner is trained for k rounds, each of the k training sets of equal size is used as a validation set in turn, and the remaining k-1 training sets are used as training sets;
[0035] Perform k rounds of training on each base learner. In each round of training, use the current k-1 training sets to train the current base learner, and use the current verification set to verify the currently trained base learner to obtain the prediction result.
[0036] After k rounds of training for the current base learner, k prediction results are obtained;
[0037] The prediction result obtained by each base learner corresponds to a feature column, and a feature matrix is formed based on the feature columns corresponding to each base learner;
[0038] Take the feature matrix as the meta-feature matrix, and use the meta-feature matrix and the true labels of the training set to train the meta-learner;
[0039] The water quality prediction using the trained remote sensing water quality parameter inversion model includes:
[0040] Input the water surface image data captured by the shore-based multi-spectral camera into each base learner to obtain the prediction results; then input the prediction results into the meta-learner to obtain the final prediction results.
[0041] Optionally, in the fourth implementation manner of the first aspect of the present invention, the potential pollution source risk extraction model includes:
[0042] The method of using the multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value includes:
[0043]
[0044] Among them, A represents the area of the target area; n represents the number of potential pollution source points in the target area; d represents the expected value; k ij represents the weight value; when L(d) < d, the research object is evenly distributed; when L(d) > d, the research object is agglomerated; when L(d) = d, the research object is randomly distributed;
[0045] The method of using the kernel density analysis method to obtain the density area map and change trend affected by potential pollution source risk points includes:
[0046]
[0047] Among them, n represents the total number of nature reserve samples within the known bandwidth range; l represents the bandwidth value; K represents the kernel density function; d(x1,x i ) represents the Euclidean distance value between points.
[0048] Optionally, in the fifth implementation manner of the first aspect of the present invention, the method of taking the identification of black and odorous water bodies, the identification of eutrophic water bodies, the inversion of remote sensing water quality parameters, and the extraction of potential pollution source risks as evaluation indicators, determining the weights of each evaluation indicator, and constructing a comprehensive evaluation model for black and odorous water bodies includes:
[0049] Perform data standardization processing on each evaluation indicator to obtain each processed evaluation indicator;
[0050] Calculate the information entropy of each processed evaluation indicator respectively;
[0051]
[0052] Among them, p iRepresents the standardized data of the i-th evaluation index;
[0053] The weight of each evaluation index is obtained according to the information entropy of each evaluation index;
[0054]
[0055] Among them, w i represents the weight of the i-th evaluation index; H(X i ) represents the information entropy of the ith evaluation index; m represents the total number of evaluation indicators.
[0056] Optionally, in a sixth implementation of the first aspect of the present invention, the method further comprises: evaluating the accuracy of the comprehensive evaluation model of black and odorous water bodies using the overall recognition accuracy ORA and the Kappa coefficient;
[0057]
[0058] Among them, CN sum Indicates the number of samples correctly identified in the test set, including normal water samples and black and smelly water samples; N test Indicates the total amount of test samples;
[0059] The Kappa coefficient includes:
[0060]
[0061] Among them, RA BOW and RA ORW are the overall recognition accuracy of black and odorous water samples and normal water samples; CN BOW and CN ORW are the number of samples correctly identified as black and odorous water bodies and the number of samples of normal water bodies in the test set; N BOW and N ORW are the real number of black and smelly water samples and the number of normal water samples in the test set samples, respectively.
[0062] The second aspect of the present invention provides a black and odorous water body remote sensing comprehensive analysis system, comprising:
[0063] An evaluation index acquisition module is used to identify black and odorous water bodies, identify eutrophic water bodies, invert remote sensing water quality parameters, and extract potential pollution source risks based on the black and odorous water body identification model, the eutrophic water body identification model, the remote sensing water quality parameter inversion model, and the potential pollution source risk extraction model;
[0064] The black and odorous water body comprehensive evaluation model construction module is used to use the black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction as evaluation indicators, determine the weight of each evaluation indicator, and construct a black and odorous water body comprehensive evaluation model;
[0065] A black and odorous water body identification module is used to identify black and odorous water bodies using the constructed black and odorous water body comprehensive evaluation model;
[0066] The black and odorous water body identification model realizes the black and odorous water body identification by combining different black and odorous index models with on-site measured spectral information;
[0067] The eutrophication water body identification model is based on lake bloom distribution and water eutrophication degree to achieve eutrophication water body identification;
[0068] The remote sensing water quality parameter inversion model uses the prediction results of different types of base learners as the input of the meta-learner to achieve water quality prediction;
[0069] The potential pollution source risk extraction model adopts a multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value, and uses the kernel density analysis method to obtain the density area map and change trend of the potential pollution source risk points.
[0070] Optionally, in a first implementation of the second aspect of the present invention, the black and odorous water body identification model includes:
[0071] The black and odorous water bodies are initially extracted based on the black and odorous water index empirical threshold, including the black and odorous water body difference index model, the normalized difference black and odorous water body index model and the black and odorous water body index model, combined with historical spectral information;
[0072] The black and odorous water body difference index model includes:
[0073] DBWI=R rs (green)-R rs (blue)
[0074] The normalized difference black and odorous water index model includes:
[0075]
[0076] The black and odorous water index model includes:
[0077]
[0078] Among them, R rs Represents the reflectivity value of the corresponding band after atmospheric correction; R rs (green) represents the reflectance value of the green band after atmospheric correction; Rrs (blue) indicates the reflectance value of the blue band after atmospheric correction; R rs (red) indicates the reflectance value of the red band after atmospheric correction;
[0079] Based on the preliminary extraction of black and odorous water bodies and combined with field measured data, the characteristic threshold of the black and odorous index is updated to accurately extract black and odorous water bodies.
[0080] Optionally, in a second implementation of the second aspect of the present invention, the eutrophic water body identification model includes:
[0081] Using the Normalized Difference Vegetation Index (NDVI) to identify the distribution of algal blooms in lakes:
[0082]
[0083] Among them, R NIR Represents the spectral reflectance value in the near-infrared band; R RED Indicates the spectral reflectance value of the red band;
[0084] The comprehensive nutrient status index TLI (∑) is used to evaluate the eutrophication degree of water bodies:
[0085]
[0086] Wherein, TLI(j) represents the nutritional status parameter of the jth parameter; W ij represents the relevant weight of the nutritional status index of the jth parameter; r ij represents the correlation coefficient between the jth parameter and the benchmark parameter total phosphorus; m represents the number of evaluation parameters.
[0087] Optionally, in a third implementation of the second aspect of the present invention, the remote sensing water quality parameter inversion model includes:
[0088] Use the water quality spectrum library to train the remote sensing water quality parameter inversion model, and use the trained remote sensing water quality parameter inversion model to predict water quality;
[0089] The method of using a water quality spectrum library to train a remote sensing water quality parameter inversion model comprises:
[0090] Based on the water quality spectral library, k training sets of equal size were constructed using the k-fold cross-validation method;
[0091] When each base learner is trained for k rounds, each of the k training sets of equal size is used as a validation set in turn, and the remaining k-1 training sets are used as training sets;
[0092] Each base learner is trained for k rounds. In each round of training, the current base learner is trained using the current k - 1 training sets, and the current trained base learner is verified using the current validation set to obtain prediction results;
[0093] After training the current base learner for k rounds, k prediction results are obtained;
[0094] The prediction results obtained by each base learner are corresponded to a feature column, and a feature matrix is formed based on the feature columns corresponding to each base learner;
[0095] The feature matrix is used as the meta - feature matrix, and the meta - learner is trained using the meta - feature matrix and the true labels of the training set;
[0096] The water quality prediction using the trained remote - sensing water quality parameter inversion model includes:
[0097] The water surface image data captured by the shore - based multi - spectral camera is input into each base learner to obtain prediction results; then the prediction results are input into the meta - learner to obtain the final prediction results.
[0098] Optionally, in the fourth implementation manner of the second aspect of the present invention, the potential pollution source risk extraction model includes:
[0099] The use of the multi - distance spatial clustering analysis method to analyze the spatial aggregation characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value includes:
[0100]
[0101] Among them, A represents the area of the target region; n represents the number of potential pollution source points in the target region; d represents the expected value; k ij represents the weight value; when L(d) < d, the research object is evenly distributed; when L(d) > d, the research object is aggregated; when L(d) = d, the research object is randomly distributed;
[0102] The use of the kernel density analysis method to obtain the density region map and change trend affected by potential pollution source risk points includes:
[0103]
[0104] Among them, n represents the total number of nature reserve samples within the known bandwidth range; l represents the bandwidth value; K represents the kernel density function; d(x1, x i ) represents the Euclidean distance value between points.
[0105] Optionally, in the fifth implementation manner of the second aspect of the present invention, the black and odorous water body comprehensive evaluation model construction module includes:
[0106] Perform data standardization on each evaluation index to obtain each evaluation index after processing;
[0107] Calculate the information entropy of each evaluation index after processing;
[0108]
[0109] Among them, p i Represents the standardized data of the i-th evaluation index;
[0110] The weight of each evaluation index is obtained according to the information entropy of each evaluation index;
[0111]
[0112] Among them, w i represents the weight of the i-th evaluation index; H(X i ) represents the information entropy of the ith evaluation index; m represents the total number of evaluation indicators.
[0113] Optionally, in a sixth implementation of the second aspect of the present invention, the device further comprises: an evaluation module for evaluating the accuracy of the comprehensive evaluation model of black and odorous water bodies using the overall recognition accuracy ORA and the Kappa coefficient;
[0114]
[0115] Among them, CN sum Indicates the number of samples correctly identified in the test set, including normal water samples and black and smelly water samples; N test Indicates the total amount of test samples;
[0116] The Kappa coefficient includes:
[0117]
[0118] Among them, RA BOW and RA ORW are the overall recognition accuracy of black and odorous water samples and normal water samples respectively; CN BOW and CN ORW are the number of samples correctly identified as black and odorous water bodies and the number of samples of normal water bodies in the test set; N BOW and N ORW are the real number of black and smelly water samples and the number of normal water samples in the test set samples, respectively.
[0119] The third aspect of the present invention provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, the steps of the above-mentioned method for comprehensive remote sensing analysis of black and odorous water bodies are implemented.
[0120] A fourth aspect of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the above-mentioned method for comprehensive remote sensing analysis of black and odorous water bodies.
[0121] Compared with the prior art, the present invention has the following beneficial effects:
[0122] 1. The present invention significantly improves the overall versatility and accuracy by integrating multiple data sources and indicators; it not only combines multiple black and odorous water identification indexes, including: normalized difference black and odorous water index (NDBWI), difference black and odorous water index (DBWI), etc., but also introduces the eutrophication water index to consider the impact of eutrophication water on the formation of black and odorous water; in addition, the application of remote sensing water quality quantitative inversion results provides more refined data support for water quality evaluation; it also considers potential risk factors, including pollution source data of POI (point of interest) and AOI (analysis area), which helps to identify potential pollution sources that may lead to the formation of black and odorous water, thereby further improving the overall accuracy and practicality.
[0123] 2. The present invention constructs a comprehensive and accurate black and odorous water body evaluation system by integrating these diverse data and indicators, adopts the entropy method to determine the weight of each evaluation indicator, conducts comprehensive analysis on the water body, extracts the black and odorous water body points, improves portability, and ensures the objectivity and scientificity of the evaluation results.
[0124] 3. The present invention can be applied to different geographical regions and water environments, providing strong technical support for water environment management, pollution prevention and control, and ecological restoration. BRIEF DESCRIPTION OF THE DRAWINGS
[0125] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:
[0126] Figure 1 A first flow chart of the method for comprehensive remote sensing analysis of black and odorous water bodies provided in an embodiment of the present invention.
[0127] Figure 2 A second flow chart of the black and odorous water body remote sensing comprehensive analysis method provided in an embodiment of the present invention.
[0128] Figure 3 A schematic structural diagram of a black and odorous water body remote sensing comprehensive analysis device provided in an embodiment of the present invention.
[0129] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0130] The embodiment of the present invention provides a method, device, equipment and storage medium for comprehensive remote sensing assessment of black and odorous water bodies, including: based on a black and odorous water body identification model, a eutrophication water body identification model, a remote sensing water quality parameter inversion model and a potential pollution source risk extraction model, respectively performing black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction; using black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction as evaluation indicators, determining the weight of each evaluation indicator, and constructing a comprehensive evaluation model for black and odorous water bodies; using the constructed comprehensive evaluation model for black and odorous water bodies to identify black and odorous water bodies ; Among them, the black and odorous water body identification model realizes the black and odorous water body identification by combining different black and odorous index models with the field measured spectral information; the eutrophication water body identification model realizes the eutrophication water body identification based on the lake body eutrophication distribution and the eutrophication degree of the water body; the remote sensing water quality parameter inversion model uses the prediction results of different types of base learners as the input of the meta-learner to realize water quality prediction; the potential pollution source risk extraction model adopts the multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of the potential pollution source risk points and determine the kernel density analysis bandwidth value, and uses the kernel density analysis method to obtain the density area map and change trend of the potential pollution source risk points. The present invention solves the technical problems that the black and odorous water body identification evaluation index is single in the prior art, the black and odorous water body identification is not comprehensive and accurate, and at the same time, there is a lack of research and judgment on potential risk factors.
[0131] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0132] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 The first embodiment of the remote sensing comprehensive judgment method for black and odorous water bodies in the embodiment of the present invention includes:
[0133] 101. Based on the black and odorous water body identification model, eutrophication water body identification model, remote sensing water quality parameter inversion model and potential pollution source risk extraction model, black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction are performed respectively;
[0134] In this embodiment, the black and odorous water body identification model includes a black and odorous water body difference index model, a normalized difference black and odorous water body index model and a black and odorous water body index model, and combines historical spectral information to perform preliminary extraction of black and odorous water bodies according to the black and odorous index empirical threshold; the different reflectivity characteristics of black and odorous water bodies and general water bodies are used to enhance the spectral difference between the two. In the red to green light band, the change amplitude of general water bodies is more obvious than that of black and odorous water bodies;
[0135] The black and odorous water difference index model includes:
[0136] DBWI=R rs (green)-R rs (blue)
[0137] The normalized difference black and odorous water index model includes:
[0138]
[0139] The black and odorous water index model includes:
[0140]
[0141] Among them, R rs Represents the reflectivity value of the corresponding band after atmospheric correction; R rs (green) represents the reflectance value of the green band after atmospheric correction; R rs (blue) indicates the reflectance value of the blue band after atmospheric correction; R rs (red) indicates the reflectance value of the red band after atmospheric correction;
[0142] Based on the preliminary extraction of black and odorous water bodies and combined with field measured data, the characteristic threshold of the black and odorous index is updated to accurately extract black and odorous water bodies.
[0143] In this embodiment, the results show that the ground spectral reflectance of black and smelly water bodies, whether gray-green or black-gray, changes more slowly than that of ordinary water bodies. In the range of 400-500nm, the reflectance value of black and smelly water bodies rises slowly; in the band of 500-700nm, normal water bodies have an obvious fluorescence peak near 550nm and 700nm, and an obvious absorption valley near 670nm, and black and smelly water bodies show a larger peak width and lower value; in the band of 750-900nm, the spectral reflectance of black and smelly water bodies will be significantly higher than that of ordinary water bodies. According to the above rules, a three-black and smelly index division model is established to accurately extract and divide the monitoring area by remote sensing.
[0144] In this embodiment, the eutrophic water body identification model includes:
[0145] Identifying the distribution of lake algal blooms using the Normalized Difference Vegetation Index NDVI:
[0146]
[0147] Among them, R NIR represents the spectral reflectance value in the near-infrared band; R RED represents the spectral reflectance value in the red band;
[0148] The comprehensive trophic state index TLI(∑) is used to evaluate the eutrophication degree of water bodies:
[0149] In this embodiment, three indicators, namely TP, TN, and CODMn, which are directly related to eutrophication, are selected as the evaluation factors for the eutrophication degree of water bodies;
[0150]
[0151] Among them, TLI(j) represents the trophic state parameter of the jth parameter; W ij represents the relevant weight of the trophic state index of the jth parameter; the values are: W TN = 0.0718, W TP = 0.1370, W IMN = 0.1840. Taking total phosphorus as the reference parameter; r ij represents the correlation coefficient between the jth parameter and the reference parameter total phosphorus; m represents the number of evaluation parameters.
[0152] A series of continuous numbers from 0 to 100 are used to classify the water body trophic state. TLI(Σ) < 30 is oligotrophic, 30 ≤ TLI(Σ) ≤ 50 is mesotrophic, 50 < TLI(Σ) ≤ 60 is slightly eutrophic, 60 < TLI(Σ) ≤ 70 is moderately eutrophic, and TLI(Σ) > 70 is severely eutrophic.
[0153] The calculation formula for each trophic state index TLI(j) is:
[0154]
[0155] Among them, ρ TN is the total nitrogen concentration, with the unit of mg / L; ρ TP is the total phosphorus concentration, with the unit of mg / L; ρ IMN is the permanganate index concentration, with the unit of mg / L.
[0156] In this embodiment, the remote sensing water quality parameter inversion model includes:
[0157] Collect the spectral curves of different types of water bodies through a fiber optic spectrometer, and construct a water quality spectral library corresponding to the fiber optic spectrometer according to the spectral curves of different types of water bodies and the measured values of the corresponding water quality parameters;
[0158] In this embodiment, in order to ensure the uniformity of the water type measured at the sampling point, field sampling is performed in equal proportions for different water quality categories. During the appropriate sampling period, a fiber optic spectrometer is used to measure the dark current spectrum curve, sky light spectrum curve, gray plate spectrum curve and water spectrum curve of the water sample point.
[0159] Based on the above data, the water-off-water radiation brightness and remote sensing reflectivity of the water body are calculated. The specific calculation formula is as follows:
[0160] S w =S sw -rS sky
[0161] Where: S w Indicates the water-leaving radiation brightness of the water body; S sw represents the spectral measurement data of the water body; r represents the wind speed; r generally takes a value between 0.021 and 0.05, wherein the calm water surface takes a value of 0.022, the 5m / s wind speed takes a value of 0.025, and the 10m / s wind speed takes a value of 0.026 to 0.028; in the present invention, the spectral sampling is generally for the calm water surface, so the value is 0.022; S sky Represents spectral measurements of sky light;
[0162]
[0163] Where: R rs Represents the remote sensing reflectivity of water body; S w represents the water-leaving radiance of the water body; ρ p represents the reflectivity of the gray plate; in this embodiment, the reflectivity of the gray plate is 0.3; S p Represents the spectral measurement data of the gray card.
[0164] According to the calculated remote sensing reflectance curve of the water sample point, the data that does not meet the standard water body spectral characteristics are screened out.
[0165] In this embodiment, water quality sampling and spectral sampling are carried out simultaneously. Sampling is done at 0.5m or 1 / 3 below the water surface. All water sample bottles are stored at low temperature and away from light. After recording the sampling site data, the water quality test results are sent to an authoritative laboratory to test the water quality parameter results, and the actual measured values of permanganate index, ammonia nitrogen and total phosphorus at the corresponding water sampling points are obtained.
[0166] According to the test results of water quality sampling, the water spectral data are classified into six different water quality categories: Class I, Class II, Class III, Class IV, Class V, and Class V-minus. At the same time, the water quality parameter concentration value corresponding to each spectral curve is recorded, and finally a water quality spectral library corresponding to the fiber optic spectrometer is constructed.
[0167] In this embodiment, the spectral curves and measured water quality parameters of different types of water bodies are collected by the fiber optic spectrometer, and a water quality spectrum library corresponding to the fiber optic spectrometer is established, which can provide benchmark data for subsequent water quality monitoring and improve the precision and accuracy of water quality monitoring.
[0168] Perform equivalent spectrum conversion on the water quality spectrum library corresponding to the current fiber optic spectrometer, and build the water quality spectrum library corresponding to the shore-based multi-spectral camera;
[0169] In this embodiment, there is an obvious scale effect between the water body remote sensing reflectance calculated by the fiber optic spectrometer and the shore-based multispectral data. Therefore, the remote sensing reflectance of the fiber optic spectrometer is resampled to the central wavelength of the shore-based multispectral image according to the spectral response function of the shore-based multispectral data to obtain an equivalent remote sensing reflectance applicable to the shore-based multispectral data. The specific calculation formula is as follows:
[0170]
[0171] Among them, R rs (band i ) represents the equivalent remote sensing reflectivity of the i-band of the shore-based multispectral camera; λ1 and λ2 are the band ranges of the i-band; R rs (λ) represents the water remote sensing reflectance of the fiber optic spectrometer; SRF(λ) represents the spectral response function at wavelength λ.
[0172] A training set is constructed based on a water quality spectral library suitable for shore-based multispectral cameras, and a remote sensing water quality parameter inversion model is trained using the training set.
[0173] In this embodiment, the water quality spectrum library corresponding to the shore-based multispectral camera is divided into a training set and a test set; the training set is used to train the remote sensing water quality parameter inversion model; the test set is used to evaluate the performance of the trained remote sensing water quality parameter inversion model;
[0174] In this embodiment, based on the water quality spectral library suitable for shore-based multispectral cameras, a 5-fold cross-validation method is used to construct 5 training sets of equal size;
[0175] When each base learner is trained for 5 rounds, each of the 5 training sets of equal size is used as a validation set in turn, and the remaining 4 training sets are used as training sets;
[0176] Wherein, the method of using the training set to train the remote sensing water quality parameter inversion model includes:
[0177] Each base learner is trained for 5 rounds. In each round of training, the current base learner is trained using the current 4 training sets, and the current validation set is used to validate the currently trained base learner to obtain the prediction result.
[0178] After 5 rounds of training for the current base learner, 5 prediction results are obtained;
[0179] The prediction result obtained by each base learner corresponds to a feature column, and a feature matrix is formed based on the feature columns corresponding to each base learner;
[0180] The feature matrix is used as the meta-feature matrix, and the meta-learner is trained using the meta-feature matrix and the true labels of the training set.
[0181] In this embodiment, the base learner includes: a decision tree, a random forest, a gradient boosted decision tree or an extreme random tree;
[0182] The decision tree (DT) is a supervised learning algorithm based on a tree structure, which is used for classification and regression problems. It divides the data into different decision paths and finally obtains a tree structure of decision rules. Each node represents a feature, each branch represents a decision, and the leaf node represents an output result. The decision tree is easy to understand and explain, and is suitable for modeling complex relationships.
[0183] The Random Forest (RF) is an ensemble learning algorithm that builds a powerful model based on multiple decision trees. Each tree is built based on a different random subsample and feature subset. By combining the opinions of each tree, the random forest improves the accuracy and robustness of the model, and is particularly effective for large-scale data and high-dimensional features.
[0184] The Gradient Boosting Decision Tree (GBDT) is an iterative ensemble learning algorithm that improves model performance by training multiple decision trees. It gradually improves the model and builds new trees by learning the residuals of the previous iteration. Gradient boosting decision trees can usually achieve higher accuracy and are effective in modeling complex nonlinear relationships.
[0185] The Extremely Randomized Tree (ET) is a variant of the random forest that is more random in the node splitting of the decision tree. Unlike the random forest, it not only randomly selects features, but also randomizes the selection process of the split points. This additional randomness provides greater diversity and can usually achieve better performance on some data sets.
[0186] The meta-learner is usually a simple machine learning model, such as linear regression or logistic regression.
[0187] The test set is used to evaluate the performance of the trained remote sensing water quality parameter inversion model, including:
[0188] The test set is input into the trained remote sensing water quality parameter inversion model to obtain the prediction results;
[0189] Compare the current prediction results with the true labels of the test set and calculate the evaluation indicators;
[0190] The performance of the trained remote sensing water quality parameter inversion model is evaluated according to the evaluation indicators, and the generalization ability and prediction accuracy of the current remote sensing water quality parameter inversion model are judged according to the performance of the remote sensing water quality parameter inversion model.
[0191] Collect water spectral data through shore-based multispectral cameras and pre-process the current water spectral data;
[0192] In this embodiment, the preprocessing of the current water body spectrum data includes: noise removal processing, gray plate correction processing and band synthesis processing;
[0193] The noise removal process removes background value noise from the original gray plate and water body spectral image through band calculation to obtain the gray plate and water body spectral image after noise removal.
[0194] The gray plate correction process selects several regions of interest on the processed gray plate image, counts the light intensity values of the ROI bands of the regions of interest, and selects the average value of each band for subsequent gray plate correction; performs band calculation on each band of the processed water body image to obtain a single-band water body spectral reflectance image, and the calculation formula is as follows:
[0195]
[0196] Where: R i Represents a single-band water spectral reflectance image; R wi represents the illumination intensity value of the i-th band of the water body image after noise removal; ρ p Represents the reflectivity of the gray plate; R hi It represents the average light intensity of the i-th band of the gray-board image after noise removal;
[0197] The band synthesis processing performs band synthesis on all processed single-band water body spectral reflectance images to obtain a full-band water body spectral reflectance image.
[0198] The preprocessed water spectral data are input into the trained remote sensing water quality parameter inversion model to obtain the water quality prediction results.
[0199] The water surface image data taken by the shore-based multispectral camera is input into each base learner to obtain the prediction result; the prediction result is then input into the meta learner to obtain the final prediction result.
[0200] In this embodiment, the potential pollution source risk extraction model includes: for the analysis of the pollution sources of black and odorous water bodies, the point-of-interest (POI) and area-of-interest (AOI) data are used to obtain and screen the information of residential communities and water-related factory sites on both sides of the river. The multi-distance spatial clustering analysis method is adopted to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value. The kernel density analysis method is used to obtain the density area map and change trend affected by potential pollution source risk points.
[0201] Among them, the multi-distance spatial clustering analysis is a common method for analyzing the spatial pattern of event points based on Ripley's K function. This method can statistically analyze the spatial correlation within a certain distance range to illustrate the agglomeration characteristics of feature points in the spatial scale range and multi-scale range. The multi-distance spatial clustering analysis method is used to analyze the spatial agglomeration characteristics of potential pollution sources in black and odorous water bodies and determine the kernel density analysis bandwidth value. The specific calculation formula is as follows:
[0202]
[0203] In the formula: A is the area of the target area, n is the number of potential pollution source points in the target area, d is the expected value, and k ij is the weight value. When L(d) < d, it means that the research object is evenly distributed; when L(d) > d, it means that the research object is agglomerated; when L(d) = d, it means that the research object is randomly distributed.
[0204] Among them, the kernel density analysis method is a spatial analysis method widely used to analyze data aggregation, and is often used to study the distribution quantity, distribution orientation, and agglomeration distribution area of point elements in a certain area. The core idea is that the probability of geographical elements occurring in areas with high spatial distribution density is greater than that in areas with low spatial distribution density, and the aggregation of experimental data is calculated and estimated through sample element data. With the help of the kernel density analysis method, the low-density and high-density areas of potential pollution source risks are made, and the change trend of risk points in the area is analyzed. The specific calculation formula is as follows:
[0205]
[0206] In the formula: n represents the total number of nature reserve samples within the known bandwidth range; l is the bandwidth value; K is the kernel density function; d(x1, x i ) is the Euclidean distance value between points.
[0207] 102. Taking the identification of black and odorous water bodies, the identification of eutrophic water bodies, the inversion of remote sensing water quality parameters, and the extraction of potential pollution source risks as evaluation indicators, determine the weights of each evaluation indicator, and construct a comprehensive evaluation model for black and odorous water bodies;
[0208] In this embodiment, a single evaluation index is often difficult to accurately reflect the water quality of different types of water bodies. The construction of the comprehensive index of black and odorous water bodies requires the determination of the weights of each evaluation index. The entropy method can calculate the information entropy of each index to determine its relative importance, and the weight of each index is determined by the result of the entropy method, thereby avoiding the deviation of subjective evaluation and expert judgment, and ensuring the objectivity and scientificity of the comprehensive evaluation model of black and odorous water bodies.
[0209] Perform data standardization on each evaluation index to obtain each evaluation index after processing;
[0210] Calculate the information entropy of each evaluation index after processing;
[0211]
[0212] Among them, p i Represents the standardized data of the i-th evaluation index;
[0213] The weight of each evaluation index is obtained according to the information entropy of each evaluation index;
[0214]
[0215] Among them, w i represents the weight of the i-th evaluation index; H(X i ) represents the information entropy of the ith evaluation index; m represents the total number of evaluation indicators.
[0216] 103. Use the constructed comprehensive evaluation model of black and odorous water bodies to identify black and odorous water bodies;
[0217] 104. The accuracy of the comprehensive evaluation model of black and odorous water bodies was evaluated using the overall recognition accuracy ORA and Kappa coefficient;
[0218] In this embodiment,
[0219] Among them, CN sum Indicates the number of samples correctly identified in the test set, including normal water samples and black and smelly water samples; N test Indicates the total amount of test samples;
[0220] The Kappa coefficient includes:
[0221]
[0222] Among them, RA BOW and RA ORW are the overall recognition accuracy of black and odorous water samples and normal water samples; CN BOW and CN ORWare the number of samples correctly identified as black and odorous water bodies and the number of samples of normal water bodies in the test set; N BOW and N ORW are the real number of black and smelly water samples and the number of normal water samples in the test set samples, respectively.
[0223] See also Figure 2 The second embodiment of the remote sensing comprehensive analysis method of black and odorous water bodies in the embodiment of the present invention includes:
[0224] 201. Acquire multi-source remote sensing data, and pre-process the acquired multi-source remote sensing data;
[0225] In this embodiment, acquiring multi-source remote sensing data includes acquiring satellite remote sensing data, acquiring spectral information of ground water samples, collecting ground water samples, and acquiring risk data of potential pollution sources based on POI and AOI;
[0226] The satellite remote sensing data is obtained, and the obtained satellite remote sensing data is subjected to radiation calibration, atmospheric correction and registration processing, including:
[0227] In the process of extracting remote sensing satellite image information, the original image data is the brightness value of its pixel, that is, the gray value (Digital Number, DN) that records the ground object information. It is a quantized, unitless integer value. Therefore, before conducting the experiment, it is necessary to convert the DN value of the image multispectral band pixel into the required reflectivity value. In addition, the radiation reflected or emitted by the ground object enters the satellite sensor, and is affected by various natural atmospheric factors such as atmospheric effects, solar angles, surface shapes, and the performance characteristics of the satellite sensor itself, resulting in distortion of the target ground object information obtained by the satellite sensor. Therefore, it is necessary to perform various preprocessing operations on the experimental data to achieve the purpose of correction and elimination. Different types of satellite sensors have different external conditions such as atmospheric interference, solar incidence angle, terrain undulations, etc. at the time of acquiring the image, as well as differences in the sensors themselves. Therefore, different types of remote sensing images need to be subjected to corresponding radiation correction, that is, radiation calibration and atmospheric correction. Since the present invention involves the fusion calculation of different sensor data, it is also necessary to perform preprocessing operations such as registration on the corresponding experimental data. After the above operations, the target image data required by the present invention can be obtained.
[0228] In this embodiment, radiation calibration is also called sensor calibration. The purpose of calibration is to eliminate or weaken the radiation distortion caused by the sensor itself. Sensor calibration is the basic premise for quantification of remote sensing data. Its accuracy determines the application level of remote sensing data to a large extent. The calibration process is to establish a quantitative relationship between the numerical quantization value (DN value) of the signal output by each detection element of the sensor and the actual radiation brightness value of the object in the corresponding pixel of the detector. In general, the conversion between DN value and radiation brightness can be achieved through the following linear relationship:
[0229] L=Bias+Gain×DN
[0230] Among them, L represents the radiant brightness after calibration; Bias and Gain represent the offset and gain coefficients respectively, which can be obtained from the header file of the image; DN is the grayscale value of the image pixel mentioned above.
[0231] In the present invention, the radiation calibration is mainly performed by the RadiometricCalibration tool of ENVI / IDL software, and the radiation calibration of part of the data is implemented in IDL.
[0232] In this embodiment, when electromagnetic waves pass through the atmosphere, the atmosphere not only changes the direction of light, but also affects the radiation characteristics of the remote sensing image. The influence of the atmosphere mainly refers to the absorption and scattering of sunlight and radiation from the target by the atmosphere. The process of eliminating the influence of the atmosphere is called atmospheric correction. There are two main types of atmospheric correction. One is atmospheric correction based on the radiation transfer equation, including the 6S model, LOWTRAN model, MORTARN model and ATCOR model, etc.; the other is atmospheric correction based on ground site data or auxiliary data, that is, while remote sensing imaging, the reflectivity of the imaging target is obtained synchronously, or by pre-setting a target with a known reflectivity, the ground truth data is compared with the output data of the sensor to eliminate the influence of the atmosphere.
[0233] In the present invention, the FLAASH model in ENVI / IDL software is used to perform atmospheric correction on remote sensing effects. The parameters required for atmospheric correction, such as image center coordinates, imaging time, imaging time, water vapor content, aerosol model, altitude, etc., can be obtained from the image header file or other data.
[0234] In this embodiment, image registration is a basic task in image processing, which refers to the best matching of two or more images of the same object or scene from different times, different sensors, and different perspectives. The experimental data in this article has a mismatch between the image data of different sensor types, and the corresponding pixels in the two images are offset.
[0235] In this embodiment, a longitude and latitude comparison method is used to calculate the longitude and latitude of the center of each pixel of the low-precision data based on high spatial resolution image data, and project them into the corresponding high-precision data pixels to achieve corresponding pixel matching.
[0236] In this embodiment, obtaining the spectrum information of the ground water sample includes:
[0237] In order to ensure the diversity and representativeness of water spectral library data, the layout of water spectral data sampling points needs to follow three principles, namely, time coverage, category coverage and scene coverage.
[0238] Time coverage: Due to the complex optical properties of inland water bodies, they vary greatly with the seasons. The content of inorganic salt components such as nitrogen and phosphorus in water bodies has obvious differences in summer and autumn, so the sampling time can be divided into two groups, namely summer and autumn. For rivers with obvious seasonal differences in pollution levels, the water body spectrum sampling time division should be adjusted according to the pollution season and the non-pollution season.
[0239] Category coverage: The spectral characteristics of water bodies can reflect the relative degree and scope of water pollution. In order to make the constructed water body spectral library reflect the pollution characteristics of different categories of water bodies in the region and realize rapid water quality assessment, it is necessary to count the number of rivers from Class I to Class V in the region, and sample the number of rivers according to the proportion of each type of water body. The different categories of river water bodies are arranged according to the principle of setting a sampling point every 2 km.
[0240] Scenario coverage: The optical properties of water bodies from different pollution sources vary greatly. In order to make the constructed water body spectral library able to reflect the pollution type of water bodies, it is necessary to increase the research on the reflectance spectral characteristics of various water bodies. Therefore, the distribution area of the sampling points to be set should include national examination sections, outlets, broken rivers, drinking water source protection areas, agricultural non-point sources, domestic non-point sources and industrial non-point sources and other rivers and lakes. At the same time, a certain number of sampling points should also be set for different pollution types of water bodies in the region, such as black and smelly water bodies, eutrophic water bodies and sediment pollution.
[0241] The types of data collected into the spectral database for different types of water bodies in the region mainly include spectral reflectance data, water quality parameter data and environmental supporting variable data.
[0242] Among them, spectral reflectance data collection includes:
[0243] Spectral equipment is used to measure the surface spectral reflectance data of river water bodies of different categories and scenarios in the region in different seasons. For rivers and lakes with serious blue algae outbreaks, in addition to collecting water surface spectral data, it is necessary to simultaneously collect surface cyanobacteria spectral information. For water bodies with floating objects and phytoplankton on the surface, their spectral information needs to be collected synchronously.
[0244] The spectral data were collected in clear, cloudless, windless or breezy weather. The measurement time was 10:00–14:00 when the lighting conditions were the best and the solar altitude angle was greater than 45°. The standard whiteboard was optimized and calibrated in time according to the changing weather conditions. The oblique photogrammetry was used, and the angle between the incident surface of the sunlight and the measuring surface of the instrument (θ v ) is controlled between 90°-135°, and the angle (φ) between the plane normal and the measuring surface of the instrument is controlled between 30°-35°. Each measuring point continuously collects ground object spectral data 10 times.
[0245] Among them, water quality and color parameter data collection includes:
[0246] When collecting water surface spectral data, water samples are collected simultaneously to measure water quality and water color parameters. Water quality parameters include dissolved oxygen, chemical oxygen demand, total phosphorus and ammonia nitrogen, and water color parameters include chlorophyll a concentration and total suspended matter concentration.
[0247] Among them, the environmental supporting variable data records include:
[0248] The application of water body spectral library data requires supporting parameters. Long-term experimental accumulation, relatively stable background data, and auxiliary information at different angles, locations, and seasons can not only provide reference values for model parameters, but also serve as prior knowledge data for quantitative inversion of surface parameters. Environmental supporting variable data include the acquisition time, point latitude and longitude information, surrounding environment (land use type), and water surface conditions (turbidity, odor, etc.) information recorded synchronously when collecting water surface spectral data.
[0249] Collection of surface water samples includes:
[0250] Water quality sampling and spectral sampling are carried out simultaneously, and the water quality test results are sent to an authoritative laboratory to test the water quality parameter results. The measured water quality parameters, sampling containers, water sampling volume and monitoring and analysis methods are shown in Table 1-2:
[0251] Table 1 Classification standards for urban black and odorous water pollution
[0252]
[0253]
[0254] *When the water depth is less than 25cm, the indicator is taken as 40% of the water depth.
[0255] Table 2 Water quality index determination method
[0256]
[0257] Acquisition of potential pollution source risk data based on POI and AOI
[0258] The basic geographic data required by the present invention includes various types of land feature information such as physical features, pipelines, and outlets, specifically including land development and utilization status data, entity category POI data, entity category AOI data, outlet distribution data, pipeline network distribution data, road data and other data. Based on this data, the information on man-made features around the river can be fully grasped to provide support for risk assessment.
[0259] Land use types include seven categories: agricultural land, green land, commercial and service land, transportation land, industrial and mining land, residential land and water areas.
[0260] POI data has spatial features of geographical identification, including name, category, longitude and latitude, etc., and is technical data for spatial big data analysis. POI information is collected by data suppliers through field surveys, updated in a timely manner, and the information is rich and accurate. The POI data used in the present invention comes from map services, including 15 categories such as car services, car sales, catering services, shopping services, sports and leisure services, shopping services, life services, government agencies and social groups, medical care services, accommodation services, scenic spots, commercial residences, transportation facilities services, science and education and cultural services, and company and enterprise place name address information. The kernel density estimation method is used for POI data to calculate the density of discrete points of various types of objects in the surrounding area, so that the density of the research object presents a spatial continuous change, and the corresponding density data set is generated, which can express the distribution pattern and distribution density of POI.
[0261] The outlet distribution data classification includes crop outlets, aquaculture outlets, domestic sewage outlets, industrial wastewater outlets, rural domestic sewage outlets, ditches, rivers and ports (streams), drains, urban stormwater outlets, factory rainwater outlets, and other outlets.
[0262] Pipeline network distribution data includes rainwater pipes and sewage pipes.
[0263] 202. Construct a comprehensive evaluation system for black and odorous water bodies;
[0264] Black and odorous water bodies have the characteristics of random formation process and complex development process. The scientific construction of black and odorous water body evaluation index system is the primary condition for black and odorous water body risk assessment. The present invention constructs a comprehensive evaluation system for black and odorous water bodies with four aspects as indicators, namely black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion, and potential pollution source risk extraction.
[0265] The present invention selects the black and odor water difference index (DBWI), the normalized difference black and odor water index (NDBWI) and the black and odor water index (BOI), and establishes a black and odor water screening mechanism by combining different black and odor index models with field measured spectral information. The identification model is established for different types of black and odor water bodies through three black and odor indexes, which serve as an important data basis for the research and judgment of black and odor water bodies. The overall technical route includes two stages: initial screening of black and odor water bodies and accurate extraction of black and odor water bodies.
[0266] In the first stage, the black and odor index is initially screened. Based on the previous data collection, the black and odor water bodies are initially extracted according to the black and odor index empirical threshold. The different reflectivity characteristics of black and odor water bodies and general water bodies are used to enhance the spectral difference between the two. In the red to green light band, the change range of general water bodies is more obvious than that of black and odor water bodies. The black and odor index includes:
[0267] Black and odorous water difference index DBWI
[0268] DBWI=R rs (green)-R rs (blue)
[0269] Normalized difference black and odorous water index NDBWI
[0270]
[0271] Black and odorous water index BOI
[0272]
[0273] Among them, R rs Indicates the reflectance value of the corresponding band after atmospheric correction.
[0274] In the second stage, the black and odorous water bodies are accurately extracted, and the characteristic threshold of the black and odorous index is updated in combination with the field measured data. The results show that the ground spectral reflectance of black and odorous water bodies, whether gray-green or black-gray, is more gentle than that of ordinary water bodies. In the range of 400-500nm, the reflectance value of black and odorous water bodies rises slowly; in the band range of 500-700nm, normal water bodies have an obvious fluorescence peak near 550nm and 700nm, and an obvious absorption valley near 670nm, and black and odorous water bodies show a larger peak width and lower value; in the band range of 750-900nm, the spectral reflectance of black and odorous water bodies will be significantly higher than that of ordinary water bodies. According to the above rules, a three-black and odorous index division model is established to accurately extract and divide the monitoring area by remote sensing.
[0275] Due to the discharge of domestic, industrial and agricultural sewage, the content of inorganic substances such as nitrogen and phosphorus in river water increases, which leads to eutrophication of water bodies, massive growth of algae, excessive consumption of oxygen in the water, and then the algae die and rot in large numbers due to lack of oxygen, causing black water masses and making the water body black and smelly. Therefore, eutrophication of water bodies is a potential sign for judging black and smelly water bodies.
[0276] An important feature of eutrophication of water bodies is the proliferation of algae, especially cyanobacteria. The outbreak of cyanobacterial blooms will cause a significant increase in the chlorophyll a content in the water body, resulting in changes in the spectral characteristics of the water body. The reflectivity of blue and red light decreases; the near-infrared band has an obvious vegetation characteristic "steep slope effect". Based on these characteristics, the normalized difference vegetation index (NDVI) can be used to identify the distribution of algal blooms in lakes. The NDVI calculation formula is as follows:
[0277]
[0278] Among them, R NIR is the spectral reflectance value in the near-infrared band, R RED is the spectral reflectance value of the red band.
[0279] In addition, total phosphorus (TP) is a basic element for algae growth and an important factor in eutrophication, while total nitrogen (TN) is one of the main factors that may cause eutrophication and algal blooms in rivers, lakes and reservoirs. Currently, a large number of studies focus on the eutrophication of a single indicator, lacking an overall eutrophication evaluation of the water body. Therefore, the present invention combines multiple water quality parameters to evaluate the eutrophication status of the water body.
[0280] The comprehensive trophic status index (TLI(Σ)) was used to evaluate the eutrophication degree of water bodies. Three indicators directly related to eutrophication, TP, TN and CODMn, were selected as evaluation factors for the eutrophication degree of water bodies. The TLI(Σ) model formula is as follows:
[0281]
[0282] Where TLI(j) represents the nutritional status index of the jth parameter; W j is the relevant weight of the nutritional status index of the jth parameter, and its value is: W TN =0.0718,W TP =0.1370,W IMN =0.1840. Taking total phosphorus as the reference parameter, r ijis the correlation coefficient between the j-th parameter and the reference parameter total phosphorus; m is the number of evaluation parameters. A series of consecutive numbers from 0 to 100 are used to classify the water body nutrient status. TLI(Σ) < 30 indicates oligotrophic, 30 ≤ TLI(Σ) ≤ 50 indicates mesotrophic, 50 < TLI(Σ) ≤ 60 indicates slightly eutrophic, 60 < TLI(Σ) ≤ 70 indicates moderately eutrophic, and TLI(Σ) > 70 indicates severely eutrophic.
[0283] The calculation formula for each nutrient status index TLI(j) is as follows:
[0284]
[0285] In the formula, ρ TN is the total nitrogen concentration, with the unit of mg / L; ρ TP is the total phosphorus concentration, with the unit of mg / L; ρ IMN is the permanganate index concentration, with the unit of mg / L.
[0286] Water quality parameters are often affected by various complex factors, and a single regression model may not be able to capture these complex non-linear relationships. The ensemble learning strategy can better capture the non-linear relationships in the data by combining the prediction results of multiple models, thus improving the model's ability to model the complex relationships of water quality parameters. Through ensemble learning, different types of base models can be selected, and even various machine learning algorithms can be included. This flexibility means that when predicting water quality parameters, the most appropriate model combination can be selected according to the actual problem, so as to better meet the characteristics of different water quality data. The main steps include: splitting the dataset, training the base learners, creating meta-features, training the meta-learner, and making predictions;
[0287] Step 1, split the dataset, splitting the original dataset into a training set and a test set. The training set is used to train the base learners, while the test set is used to evaluate the performance of the final model. Usually, the K-fold cross-validation method is needed to divide the training set to prevent overfitting.
[0288] Step 2, train the base learners, using the training set data to train multiple different types of base learners, such as decision trees, support vector machines, random forests, etc. Each base learner makes predictions on the dataset to obtain the prediction results;
[0289] Step 3, create meta-features, using the prediction results of each base learner on the training set as new features to form a meta-feature matrix. The prediction result of each base learner corresponds to a feature column;
[0290] Step 4: Train the meta-learner. Use the meta-feature matrix and the true labels of the training set to train a meta-learner. This meta-learner is usually a simple machine learning model, such as linear regression, logistic regression, etc. The goal of the meta-learner is to learn how to combine the prediction results of each base learner to obtain a more accurate final prediction result;
[0291] Step 5: Make predictions. Use the trained model to input the water surface image data captured by the shore-based multispectral camera into each base learner to obtain their prediction results, and then use these prediction results as inputs to obtain the final prediction result through the meta-learner.
[0292] Residential areas, construction sites, and industrial parks on both sides of the river are all potential pollution source risks. If these wastewaters or chemical substances are directly discharged into the river without proper treatment, especially in some enclosed waters, these water bodies may lack sufficient water flow and have poor self-purification ability. The accumulation and decomposition of organic matter generated by their illegal discharge lead to the emergence of black and odorous water bodies. For the analysis of the pollution sources of black and odorous water bodies, obtain and screen the information of residential communities and water-related factory sites on both sides of the river through POI and AOI data. Use the multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value. Use the kernel density analysis method to obtain the density area map and change trend affected by potential pollution source risk points.
[0293] The multi-distance spatial clustering analysis is a common method based on Ripley's K function to analyze the spatial pattern of event points. This method can statistically analyze the spatial correlation within a certain distance range to illustrate the agglomeration characteristics of feature points in the spatial scale range and multi-scale range. Use the multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution sources of black and odorous water bodies and determine the kernel density analysis bandwidth value. The specific calculation formula is as follows:
[0294]
[0295] In the formula: A is the area of the target region, n is the number of potential pollution source points in the target region, d is the expected value, and k ij is the weight value. When L(d) < d, it means that the research object is evenly distributed; when L(d) > d, it means that the research object is agglomerated; when L(d) = d, it means that the research object is randomly distributed.
[0296] The kernel density analysis method is a spatial analysis method widely used to analyze data aggregation. It is often used to study the distribution quantity, distribution orientation, and clustered distribution area of point elements in a certain area. The core idea is that the probability of geographical elements occurring in areas with high spatial distribution density is greater than that in areas with low spatial distribution density. The clustering of experimental data is calculated and estimated through sample element data. The kernel density analysis method is used to create low-density and high-density areas of potential pollution source risks and analyze the changing trend of risk points in the area. The specific calculation formula is as follows:
[0297]
[0298] Where: n represents the total number of samples of nature reserves within a known bandwidth; l is the bandwidth value; K is the kernel density function; d(x1,x i ) is the Euclidean distance between points.
[0299] It is often difficult for a single evaluation index to accurately reflect the water quality of different types of water bodies. The construction of a comprehensive index for black and odorous water bodies requires the determination of the weights of each evaluation index. The entropy method can calculate the information entropy of each index to determine its relative importance, and determine the weight of each index through the results of the entropy method, thereby avoiding the deviation of subjective evaluation and expert judgment, and ensuring the objectivity and scientificity of the comprehensive evaluation model for black and odorous water bodies.
[0300] Data Standardization
[0301] The evaluation indicators are normalized by minimum and maximum values and converted into numerical values in the interval [0,1]. The purpose of standardization is to eliminate the dimensional differences between different factors and make them comparable.
[0302] Calculating Information Entropy
[0303] For each evaluation indicator, calculate its information entropy. Information entropy indicates the uncertainty or confusion of the data. The higher the entropy value, the greater the uncertainty. The calculation formula of entropy value is:
[0304]
[0305] Where: p i is the standardized value of the ith factor.
[0306] Calculating weights
[0307] The weights of the black and odorous water evaluation index are calculated based on information entropy. The factors with larger entropy values have smaller corresponding weights, and vice versa. The formula for calculating the weights is:
[0308]
[0309] Where: wi is the weight of the i-th factor; H(Xi) is the information entropy of the i-th factor; m is the total number of factors.
[0310] 203. Identification of black and odorous water bodies and progress evaluation;
[0311] In this embodiment, the present invention constructs a comprehensive evaluation index for black and odorous water bodies, uses the entropy method to determine the weight of each index, and establishes a comprehensive evaluation model for black and odorous water bodies to identify black and odorous water bodies. The overall recognition accuracy (ORA) and Kappa coefficient are used to evaluate the accuracy of the black and odorous water body recognition model. The overall recognition accuracy is calculated by the formula:
[0312]
[0313] In the formula, CN sum N is the number of samples correctly identified in the test set, including normal water samples and black and smelly water samples. test is the total number of test samples.
[0314] The Kappa coefficient calculation formula is as follows:
[0315]
[0316] In the formula, RA BOW and RA ORW are the overall recognition accuracy of black and odorous water samples and normal water samples, CN BOW and CN ORW are the number of samples correctly identified as black and odorous water bodies and the number of samples of normal water bodies in the test set; N BOW and N ORW are the real number of black and smelly water samples and the number of normal water samples in the test set samples, respectively.
[0317] The above describes the method for comprehensive remote sensing analysis of black and odorous water bodies in the embodiment of the present invention. The following describes the device for comprehensive remote sensing analysis of black and odorous water bodies in the embodiment of the present invention. Figure 3 In one embodiment of the present invention, a device for comprehensive remote sensing analysis of black and odorous water bodies includes:
[0318] Evaluation index acquisition module 301, used for black and odorous water body identification, eutrophic water body identification, remote sensing water quality parameter inversion model and potential pollution source risk extraction model respectively;
[0319] In this embodiment, the black and odorous water body identification model includes a black and odorous water body difference index model, a normalized difference black and odorous water body index model and a black and odorous water body index model, and combines historical spectral information to perform preliminary extraction of black and odorous water bodies according to the black and odorous index empirical threshold; the different reflectivity characteristics of black and odorous water bodies and general water bodies are used to enhance the spectral difference between the two. In the red to green light band, the change amplitude of general water bodies is more obvious than that of black and odorous water bodies;
[0320] The black and odorous water difference index model includes:
[0321] DBWI=R rs (green)-R rs (blue)
[0322] The normalized difference black and odorous water index model includes:
[0323]
[0324] The black and odorous water index model includes:
[0325]
[0326] Among them, R rs Represents the reflectivity value of the corresponding band after atmospheric correction; R rs (green) represents the reflectance value of the green band after atmospheric correction; R rs (blue) indicates the reflectance value of the blue band after atmospheric correction; R rs (red) indicates the reflectance value of the red band after atmospheric correction;
[0327] Based on the preliminary extraction of black and odorous water bodies and combined with field measured data, the characteristic threshold of the black and odorous index is updated to accurately extract black and odorous water bodies.
[0328] In this embodiment, the results show that the ground spectral reflectance of black and smelly water bodies, whether gray-green or black-gray, changes more slowly than that of ordinary water bodies. In the range of 400-500nm, the reflectance value of black and smelly water bodies rises slowly; in the band of 500-700nm, normal water bodies have an obvious fluorescence peak near 550nm and 700nm, and an obvious absorption valley near 670nm, and black and smelly water bodies show a larger peak width and lower value; in the band of 750-900nm, the spectral reflectance of black and smelly water bodies will be significantly higher than that of ordinary water bodies. According to the above rules, a three-black and smelly index division model is established to accurately extract and divide the monitoring area by remote sensing.
[0329] In this embodiment, the eutrophic water body identification model includes:
[0330] Using the Normalized Difference Vegetation Index (NDVI) to identify the distribution of algal blooms in lakes:
[0331]
[0332] Among them, R NIR represents the spectral reflectance value in the near-infrared band; R RED represents the spectral reflectance value in the red band;
[0333] The comprehensive nutrition state index TLI(∑) is used to evaluate the eutrophication degree of water bodies:
[0334] In this embodiment, three indicators of TP, TN, and CODMn, which are directly related to eutrophication, are selected as the evaluation factors for the eutrophication degree of water bodies;
[0335]
[0336] Among them, TLI(j) represents the nutrition state parameter of the jth parameter; W ij represents the relevant weight of the nutrition state index of the jth parameter; the value is: W TN = 0.0718, W TP = 0.1370, W IMN = 0.1840. Taking total phosphorus as the reference parameter; r ij represents the correlation coefficient between the jth parameter and the reference parameter total phosphorus; m represents the number of evaluation parameters.
[0337] A series of consecutive numbers from 0 to 100 are used to classify the water body nutrition state. TLI(Σ) < 30 is oligotrophic, 30 ≤ TLI(Σ) ≤ 50 is mesotrophic, 50 < TLI(Σ) ≤ 60 is slightly eutrophic, 60 < TLI(Σ) ≤ 70 is moderately eutrophic, and TLI(Σ) > 70 is severely eutrophic.
[0338] The calculation formula for each nutrition state index TLI(j) is:
[0339]
[0340] Among them, ρ TN is the total nitrogen concentration, with the unit of mg / L; ρ TP is the total phosphorus concentration, with the unit of mg / L; ρ IMN is the permanganate index concentration, with the unit of mg / L.
[0341] In this embodiment, the remote sensing water quality parameter inversion model includes:
[0342] The spectral curves of different types of water bodies are collected by a fiber optic spectrometer, and according to the spectral curves of different types of water bodies and the corresponding measured values of water quality parameters, a water quality spectral library corresponding to the fiber optic spectrometer is constructed;
[0343] In this embodiment, in order to ensure the uniformity of the water type measured at the sampling point, field sampling is performed in equal proportions for different water quality categories. During the appropriate sampling period, a fiber optic spectrometer is used to measure the dark current spectrum curve, sky light spectrum curve, gray plate spectrum curve and water spectrum curve of the water sample point.
[0344] Based on the above data, the water-off-water radiation brightness and remote sensing reflectivity of the water body are calculated. The specific calculation formula is as follows:
[0345] S w =S sw -rS sky
[0346] Where: S w Indicates the water-leaving radiation brightness of the water body; S sw represents the spectral measurement data of the water body; r represents the wind speed; r generally takes a value between 0.021 and 0.05, wherein the calm water surface takes a value of 0.022, the 5m / s wind speed takes a value of 0.025, and the 10m / s wind speed takes a value of 0.026 to 0.028; in the present invention, the spectral sampling is generally for the calm water surface, so the value is 0.022; S sky Represents spectral measurements of sky light;
[0347]
[0348] Where: R rs Represents the remote sensing reflectivity of water body; S w represents the water-free radiance of the water body; ρ p represents the reflectivity of the gray plate; in this embodiment, the reflectivity of the gray plate is 0.3; S p Represents the spectral measurement data of the gray card.
[0349] According to the calculated remote sensing reflectance curve of the water sample point, the data that does not meet the standard water body spectral characteristics are screened out.
[0350] In this embodiment, water quality sampling and spectral sampling are carried out simultaneously. Sampling is done at 0.5m or 1 / 3 below the water surface. All water sample bottles are stored at low temperature and away from light. After recording the sampling site data, the water quality test results are sent to an authoritative laboratory to test the water quality parameter results, and the actual measured values of permanganate index, ammonia nitrogen and total phosphorus at the corresponding water sampling points are obtained.
[0351] According to the test results of water quality sampling, the water spectral data are classified into six different water quality categories: Class I, Class II, Class III, Class IV, Class V, and Class V-minus. At the same time, the water quality parameter concentration value corresponding to each spectral curve is recorded, and finally a water quality spectral library corresponding to the fiber optic spectrometer is constructed.
[0352] In this embodiment, the spectral curves and measured water quality parameters of different types of water bodies are collected by the fiber optic spectrometer, and a water quality spectrum library corresponding to the fiber optic spectrometer is established, which can provide benchmark data for subsequent water quality monitoring and improve the precision and accuracy of water quality monitoring.
[0353] Perform equivalent spectrum conversion on the water quality spectrum library corresponding to the current fiber optic spectrometer, and build the water quality spectrum library corresponding to the shore-based multi-spectral camera;
[0354] In this embodiment, there is an obvious scale effect between the water body remote sensing reflectance calculated by the fiber optic spectrometer and the shore-based multispectral data. Therefore, the remote sensing reflectance of the fiber optic spectrometer is resampled to the central wavelength of the shore-based multispectral image according to the spectral response function of the shore-based multispectral data to obtain an equivalent remote sensing reflectance applicable to the shore-based multispectral data. The specific calculation formula is as follows:
[0355]
[0356] Among them, R rs (band i ) represents the equivalent remote sensing reflectivity of the i-band of the shore-based multispectral camera; λ1 and λ2 are the band ranges of the i-band; R rs (λ) represents the water remote sensing reflectance of the fiber optic spectrometer; SRF(λ) represents the spectral response function at wavelength λ.
[0357] A training set is constructed based on a water quality spectral library suitable for shore-based multispectral cameras, and a remote sensing water quality parameter inversion model is trained using the training set.
[0358] In this embodiment, the water quality spectrum library corresponding to the shore-based multispectral camera is divided into a training set and a test set; the training set is used to train the remote sensing water quality parameter inversion model; the test set is used to evaluate the performance of the trained remote sensing water quality parameter inversion model;
[0359] In this embodiment, based on the water quality spectral library suitable for shore-based multispectral cameras, a 5-fold cross-validation method is used to construct 5 training sets of equal size;
[0360] When each base learner is trained for 5 rounds, each of the 5 training sets of equal size is used as a validation set in turn, and the remaining 4 training sets are used as training sets;
[0361] Wherein, the method of using the training set to train the remote sensing water quality parameter inversion model includes:
[0362] Each base learner is trained for 5 rounds. In each round of training, the current base learner is trained using the current 4 training sets, and the current validation set is used to validate the currently trained base learner to obtain the prediction result.
[0363] After 5 rounds of training for the current base learner, 5 prediction results are obtained;
[0364] The prediction result obtained by each base learner corresponds to a feature column, and a feature matrix is formed based on the feature columns corresponding to each base learner;
[0365] The feature matrix is used as the meta-feature matrix, and the meta-learner is trained using the meta-feature matrix and the true labels of the training set.
[0366] In this embodiment, the base learner includes: a decision tree, a random forest, a gradient boosted decision tree or an extreme random tree;
[0367] The decision tree (DT) is a supervised learning algorithm based on a tree structure, which is used for classification and regression problems. It divides the data into different decision paths and finally obtains a tree structure of decision rules. Each node represents a feature, each branch represents a decision, and the leaf node represents an output result. The decision tree is easy to understand and explain, and is suitable for modeling complex relationships.
[0368] The Random Forest (RF) is an ensemble learning algorithm that builds a powerful model based on multiple decision trees. Each tree is built based on a different random subsample and feature subset. By combining the opinions of each tree, the random forest improves the accuracy and robustness of the model, and is particularly effective for large-scale data and high-dimensional features.
[0369] The Gradient Boosting Decision Tree (GBDT) is an iterative ensemble learning algorithm that improves model performance by training multiple decision trees. It gradually improves the model and builds new trees by learning the residuals of the previous iteration. Gradient boosting decision trees can usually achieve higher accuracy and are effective in modeling complex nonlinear relationships.
[0370] The Extremely Randomized Tree (ET) is a variant of the random forest that is more random in the node splitting of the decision tree. Unlike the random forest, it not only randomly selects features, but also randomizes the selection process of the split points. This additional randomness provides greater diversity and can usually achieve better performance on some data sets.
[0371] The meta-learner is usually a simple machine learning model, such as linear regression or logistic regression.
[0372] The test set is used to evaluate the performance of the trained remote sensing water quality parameter inversion model, including:
[0373] The test set is input into the trained remote sensing water quality parameter inversion model to obtain the prediction results;
[0374] Compare the current prediction results with the true labels of the test set and calculate the evaluation indicators;
[0375] The performance of the trained remote sensing water quality parameter inversion model is evaluated according to the evaluation indicators, and the generalization ability and prediction accuracy of the current remote sensing water quality parameter inversion model are judged according to the performance of the remote sensing water quality parameter inversion model.
[0376] Collect water spectral data through shore-based multispectral cameras and pre-process the current water spectral data;
[0377] In this embodiment, the preprocessing of the current water body spectrum data includes: noise removal processing, gray plate correction processing and band synthesis processing;
[0378] The noise removal process removes background value noise from the original gray plate and water body spectral image through band calculation to obtain the gray plate and water body spectral image after noise removal.
[0379] The gray plate correction process selects several regions of interest on the processed gray plate image, counts the light intensity values of the ROI bands of the regions of interest, and selects the average value of each band for subsequent gray plate correction; performs band calculation on each band of the processed water body image to obtain a single-band water body spectral reflectance image, and the calculation formula is as follows:
[0380]
[0381] Where: R i Represents a single-band water spectral reflectance image; R wi represents the illumination intensity value of the i-th band of the water body image after noise removal; ρ p Represents the reflectivity of the gray plate; R hi It represents the average light intensity of the i-th band of the gray-board image after noise removal;
[0382] The band synthesis process performs band synthesis on all processed single-band water body spectral reflectance images to obtain a full-band water body spectral reflectance image.
[0383] The preprocessed water spectral data are input into the trained remote sensing water quality parameter inversion model to obtain the water quality prediction results.
[0384] The water surface image data taken by the shore-based multispectral camera is input into each base learner to obtain the prediction result; the prediction result is then input into the meta learner to obtain the final prediction result.
[0385] In this embodiment, the potential pollution source risk extraction model includes: for the analysis of black and odorous water body pollution sources, obtaining and screening the information of residential communities and water-related factory sites on both sides of the river through POI and AOI data, using the multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value, and using the kernel density analysis method to obtain the density area map and change trend of the influence of potential pollution source risk points.
[0386] Among them, the multi-distance spatial clustering analysis is a commonly used method for analyzing the spatial pattern of event points based on Ripley's K function. This method can statistically analyze the spatial correlation within a certain distance range to illustrate the agglomeration characteristics of feature points in the spatial scale range and multi-scale range. The multi-distance spatial clustering analysis method is used to analyze the spatial agglomeration characteristics of potential pollution sources in black and odorous water bodies and determine the kernel density analysis bandwidth value. The specific calculation formula is as follows:
[0387]
[0388] In the formula: A is the area of the target area, n is the number of potential pollution source points in the target area, d is the expected value, and k ij is the weight value. When L(d) < d, it means that the research object is evenly distributed; when L(d) > d, it means that the research object is agglomerated; when L(d) = d, it means that the research object is randomly distributed.
[0389] Among them, the kernel density analysis method is a spatial analysis method widely used to analyze data aggregation, and is often used to study the distribution quantity, distribution orientation, and agglomeration distribution area of point elements in a certain area. The core idea is that the probability of geographical elements occurring in areas with high spatial distribution density is greater than that in areas with low spatial distribution density, and the aggregation of experimental data is calculated and estimated through sample element data. With the help of the kernel density analysis method, low-density and high-density areas of potential pollution source risks are made, and the change trend of risk points in the area is analyzed. The specific calculation formula is as follows:
[0390]
[0391] In the formula: n represents the total number of nature reserve samples within the known bandwidth range; l is the bandwidth value; K is the kernel density function; d(x1, x i ) is the Euclidean distance value between points.
[0392] The black and odorous water body comprehensive evaluation model construction module 302 is used to use the black and odorous water body identification, eutrophic water body identification, remote sensing water quality parameter inversion, and potential pollution source risk extraction as evaluation indicators, determine the weights of each evaluation indicator, and construct a black and odorous water body comprehensive evaluation model;
[0393] In this embodiment, a single evaluation index is often difficult to accurately reflect the water quality of different types of water bodies. The construction of the comprehensive index of black and odorous water bodies requires the determination of the weights of each evaluation index. The entropy method can calculate the information entropy of each index to determine its relative importance, and the weight of each index is determined by the result of the entropy method, thereby avoiding the deviation of subjective evaluation and expert judgment, and ensuring the objectivity and scientificity of the comprehensive evaluation model of black and odorous water bodies.
[0394] Perform data standardization on each evaluation index to obtain each evaluation index after processing;
[0395] Calculate the information entropy of each evaluation index after processing;
[0396]
[0397] Among them, p i Represents the standardized data of the i-th evaluation index;
[0398] The weight of each evaluation indicator is obtained according to the information entropy of each evaluation indicator;
[0399]
[0400] Among them, w i represents the weight of the i-th evaluation index; H(X i ) represents the information entropy of the ith evaluation index; m represents the total number of evaluation indicators.
[0401] A black and odorous water body identification module 303 is used to identify black and odorous water bodies using the constructed black and odorous water body comprehensive evaluation model;
[0402] An evaluation module 304 is used to evaluate the accuracy of the comprehensive evaluation model of black and odorous water bodies using the overall recognition accuracy ORA and Kappa coefficient;
[0403] In this embodiment,
[0404] Among them, CN sum Indicates the number of samples correctly identified in the test set, including normal water samples and black and smelly water samples; N test Indicates the total amount of test samples;
[0405] The Kappa coefficient includes:
[0406]
[0407] Among them, RA BOW and RA ORW are the overall recognition accuracy of black and odorous water samples and normal water samples respectively; CN BOW and CNORW are the number of samples correctly identified as black and odorous water bodies and the number of samples of normal water bodies in the test set; N BOW and N ORW are the real number of black and smelly water samples and the number of normal water samples in the test set samples, respectively.
[0408] above Figure 3 The black and odorous water body remote sensing comprehensive analysis device in the embodiment of the present invention is described in detail from the perspective of modular functional entities, and the electronic device in the embodiment of the present invention is described in detail from the perspective of hardware processing.
[0409] Figure 4 7 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device 700 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 710 (for example, one or more processors) and a memory 720, and one or more storage media 730 (for example, one or more mass storage devices) storing application programs 733 or data 732. Among them, the memory 720 and the storage medium 730 can be temporary storage or permanent storage. The program stored in the storage medium 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the electronic device 700. Furthermore, the processor 710 can be configured to communicate with the storage medium 730 to execute a series of instruction operations in the storage medium 730 on the electronic device 700.
[0410] The electronic device 700 may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input and output interfaces 750, and / or one or more operating systems 731, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 4 The structure of the electronic device shown does not constitute a limitation on the electronic device, and may include more or less components than shown in the figure, or combine some components, or arrange the components differently.
[0411] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for comprehensive remote sensing analysis of black and odorous water bodies.
[0412] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.
[0413] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.
[0414] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A remote sensing comprehensive judgment method for black and odorous water bodies, characterized in that: include: Based on the black and odorous water body identification model, eutrophication water body identification model, remote sensing water quality parameter inversion model and potential pollution source risk extraction model, black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction were carried out respectively; The black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction are used as evaluation indicators, the weight of each evaluation indicator is determined, and a comprehensive evaluation model for black and odorous water bodies is constructed; The black and odorous water bodies are identified using the constructed comprehensive evaluation model of black and odorous water bodies; The black and odorous water body identification model realizes the black and odorous water body identification by combining different black and odorous index models with on-site measured spectral information; The eutrophication water body identification model is based on lake bloom distribution and water eutrophication degree to achieve eutrophication water body identification; The remote sensing water quality parameter inversion model uses the prediction results of different types of base learners as the input of the meta-learner to achieve water quality prediction; The potential pollution source risk extraction model adopts a multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value, and uses the kernel density analysis method to obtain the density area map and change trend of the potential pollution source risk points.
2. The remote sensing comprehensive judgment method for black and odorous water bodies according to claim 1 is characterized in that: The black and odorous water body identification model comprises: The black and odorous water bodies are initially extracted based on the black and odorous water index empirical threshold, including the black and odorous water body difference index model, the normalized difference black and odorous water body index model and the black and odorous water body index model, combined with historical spectral information; The black and odorous water body difference index model includes: DBWI=R rs (green)-R rs (blue) The normalized difference black and odorous water index model includes: The black and odorous water index model includes: Among them, R rs Represents the reflectivity value of the corresponding band after atmospheric correction; R rs (green) represents the reflectance value of the green band after atmospheric correction; R rs (blue) indicates the reflectance value of the blue band after atmospheric correction; R rs (red) indicates the reflectance value of the red band after atmospheric correction; Based on the preliminary extraction of black and odorous water bodies and combined with field measured data, the characteristic threshold of the black and odorous index is updated to accurately extract black and odorous water bodies.
3. The remote sensing comprehensive analysis method for black and odorous water bodies according to claim 1 is characterized in that: The eutrophic water body identification model includes: Using the Normalized Difference Vegetation Index (NDVI) to identify the distribution of algal blooms in lakes: Among them, R NIR Represents the spectral reflectance value in the near-infrared band; R RED Indicates the spectral reflectance value of the red band; The comprehensive nutrient status index TLI (∑) is used to evaluate the eutrophication degree of water bodies: Wherein, TLI(j) represents the nutritional status parameter of the jth parameter; W ij represents the relevant weight of the nutritional status index of the jth parameter; r ij represents the correlation coefficient between the jth parameter and the benchmark parameter total phosphorus; m represents the number of evaluation parameters.
4. The remote sensing comprehensive analysis method for black and odorous water bodies according to claim 1 is characterized in that: The remote sensing water quality parameter inversion model includes: Use the water quality spectrum library to train the remote sensing water quality parameter inversion model, and use the trained remote sensing water quality parameter inversion model to predict water quality; The method of using a water quality spectrum library to train a remote sensing water quality parameter inversion model comprises: Based on the water quality spectral library, k training sets of equal size were constructed using the k-fold cross-validation method; When each base learner is trained for k rounds, each of the k training sets of equal size is used as a validation set in turn, and the remaining k-1 training sets are used as training sets; Perform k rounds of training on each base learner. In each round of training, use the current k-1 training sets to train the current base learner, and use the current verification set to verify the currently trained base learner to obtain the prediction result. After k rounds of training for the current base learner, k prediction results are obtained; The prediction result obtained by each base learner corresponds to a feature column, and a feature matrix is formed based on the feature columns corresponding to each base learner; The feature matrix is used as a meta-feature matrix, and the meta-feature matrix and the true labels of the training set are used to train the meta-learner; The method of using the trained remote sensing water quality parameter inversion model to predict water quality includes: The water surface image data taken by the shore-based multispectral camera is input into each base learner to obtain the prediction result; the prediction result is then input into the meta learner to obtain the final prediction result.
5. The remote sensing comprehensive analysis method for black and odorous water bodies according to claim 1 is characterized in that: The potential pollution source risk extraction model includes: The multi-distance spatial clustering analysis method is used to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the bandwidth value of kernel density analysis, including: Among them, A represents the area of the target area; n represents the number of potential pollution source points in the target area; d represents the expected value; k ij represents the weight value; when L(d) < d, the research object shows a uniform distribution; when L(d) > d, the research object shows an agglomerated distribution; when L(d) = d, the research object shows a random distribution; The density area map and change trend of the potential pollution source risk point location obtained by the kernel density analysis method include: Where n is the total number of samples of nature reserves within the known bandwidth; l is the bandwidth value; K is the kernel density function; d(x1,x i ) represents the Euclidean distance between points.
6. The remote sensing comprehensive analysis method for black and odorous water bodies according to claim 1 is characterized in that: The black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction are used as evaluation indicators, the weight of each evaluation indicator is determined, and a comprehensive evaluation model for black and odorous water bodies is constructed, including: Perform data standardization on each evaluation index to obtain each evaluation index after processing; Calculate the information entropy of each evaluation index after processing; Among them, p i Represents the standardized data of the i-th evaluation index; The weight of each evaluation index is obtained according to the information entropy of each evaluation index; Among them, w i represents the weight of the i-th evaluation index; H(X i ) represents the information entropy of the ith evaluation index; m represents the total number of evaluation indicators.
7. The remote sensing comprehensive analysis method for black and odorous water bodies according to claim 1 is characterized in that: The method further comprises: evaluating the accuracy of the comprehensive evaluation model of black and odorous water bodies by using the overall recognition accuracy ORA and Kappa coefficient; Among them, CN sum Indicates the number of samples correctly identified in the test set, including normal water samples and black and smelly water samples; N test Indicates the total amount of test samples; The Kappa coefficient includes: Among them, RA BOW and RA ORW are the overall recognition accuracy of black and odorous water samples and normal water samples; CN BOW and CN ORW are the number of samples correctly identified as black and odorous water bodies and the number of samples of normal water bodies in the test set; N BOW and N ORW are the real number of black and smelly water samples and the number of normal water samples in the test set samples, respectively.
8. A remote sensing comprehensive analysis system for black and odorous water bodies, characterized in that: include: An evaluation index acquisition module is used to identify black and odorous water bodies, identify eutrophic water bodies, invert remote sensing water quality parameters, and extract potential pollution source risks based on the black and odorous water body identification model, the eutrophic water body identification model, the remote sensing water quality parameter inversion model, and the potential pollution source risk extraction model; The black and odorous water body comprehensive evaluation model construction module is used to use the black and odorous water body identification, eutrophication water body identification, remote sensing water quality parameter inversion and potential pollution source risk extraction as evaluation indicators, determine the weight of each evaluation indicator, and construct a black and odorous water body comprehensive evaluation model; A black and odorous water body identification module is used to identify black and odorous water bodies using the constructed black and odorous water body comprehensive evaluation model; The black and odorous water body identification model realizes the black and odorous water body identification by combining different black and odorous index models with on-site measured spectral information; The eutrophication water body identification model is based on lake bloom distribution and water eutrophication degree to achieve eutrophication water body identification; The remote sensing water quality parameter inversion model uses the prediction results of different types of base learners as the input of the meta-learner to achieve water quality prediction; The potential pollution source risk extraction model adopts a multi-distance spatial clustering analysis method to analyze the spatial agglomeration characteristics of potential pollution source risk points and determine the kernel density analysis bandwidth value, and uses the kernel density analysis method to obtain the density area map and change trend of the potential pollution source risk points.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the black and odorous water body remote sensing comprehensive analysis method described in any one of claims 1 to 7 are implemented.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of the black and odorous water body remote sensing comprehensive analysis method described in any one of claims 1 to 7 are implemented.
Citation Information
Cited By
Online water quality analysis method and system based on multispectral image analysis model
CN119555679A
Quantitative analysis method and system for spatial differentiation characteristics of water quality in water area
CN119619432A
Estuary coast land source pollution monitoring method and system
CN120121544A
River-side city black and odorous water body risk division method and system
CN120430473A
Urban water black and odor returning risk assessment system and method based on multi-interface coupling
CN120851619A
