Water area chlorophyll concentration remote sensing inversion method, system, medium and equipment

Abstract

The invention discloses a water area chlorophyll a concentration remote sensing inversion method and system, a medium and equipment, relates to the technical field of environmental monitoring, and provides an improved genetic algorithm for feature selection based on Sentinel-2 remote sensing satellite images and actually measured Chl-a concentration data of a Wuliang vegetarian sea by taking the Wuliang vegetarian sea as a research area, and an XGBoost model is optimized in combination with a dung beetle optimization algorithm DBO. And inversion is carried out on the Chl-a concentration of the ouyew vegetarian sea. A remote sensing characteristic wave band of Chl-a concentration is screened through an improved genetic algorithm, the remote sensing characteristic wave band is input into a DBO-XGBoost model, and a DBO-XGBoost Wuji prime sea Chl-a concentration inversion model is established. The result shows that under the same inversion model, redundant information between wave bands is effectively reduced, the complexity of the model is reduced, and the prediction performance of the model is improved at the same time. The decision coefficient R2 of the-DBO-XGBoost model is 0.9379, and the decision coefficient R2 of the- The root-mean-square error (RMSE) is 2.2566 [mu] g / L, the mean absolute error (MAE) is 1.9019 [mu] g / L, and powerful support is provided for monitoring of the concentration of the Wuliang vegetarian sea Chl-a.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and in particular to a remote sensing inversion method, system, medium and equipment for chlorophyll a concentration in water areas. Background Art

[0002] Lake eutrophication refers to the phenomenon that the lake water receives excessive nutrients such as phosphorus and nitrogen, which leads to the proliferation of algae and other aquatic plants, thereby consuming oxygen in the water, causing water quality deterioration and the death of fish and other organisms. Chlorophyll a is a green pigment that exists in the chloroplasts of algae, plants and some bacteria. It plays a vital role in photosynthesis, and its concentration is an important indicator for assessing the degree of algae and eutrophication in water bodies. Therefore, monitoring the concentration of chlorophyll a can effectively reflect the eutrophication of lake water bodies, which is of great significance for timely prevention and control of lake eutrophication problems and protection of lake ecological environment.

[0003] With the widespread application of remote sensing technology in the field of chlorophyll a concentration prediction in water bodies, a problem is faced: remote sensing spectral data are usually of high dimension, and there is collinearity between characteristic bands. The input of all characteristic bands maximizes the complexity of the model, and the prediction accuracy may be affected. Therefore, when establishing a chlorophyll a concentration inversion model, selecting characteristic bands that are sensitive to chlorophyll a concentration is an important prerequisite for streamlining the model, improving prediction accuracy, and enhancing generalization ability. However, it is difficult to express the relationship between remote sensing spectral features and water body elements with a simple linear relationship, so its adaptability is poor. The empirical method and analytical method are based on the combination of remote sensing data and parameters related to the inherent optical characteristics of water bodies, but they are highly dependent on the quality of spectral data and are easily affected by the concentration of other substances in the water. The inversion effect is poor when facing complex water bodies, resulting in low accuracy of prediction results. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art that the quality of spectral data is highly dependent, it is easily affected by the concentration of other substances in the water, the inversion effect is poor when facing complex water bodies, and the prediction results are not accurate, the main purpose of the present invention is to provide a remote sensing inversion method, system, medium and equipment for chlorophyll a concentration in water areas.

[0005] To achieve the above object, the present invention adopts the following technical scheme, a remote sensing inversion method for chlorophyll a concentration in water area, comprising the following steps: Acquire the chlorophyll a concentration data and remote sensing images of the target waters at the same time; pre-process the remote sensing images of the target waters to obtain the target band; Based on the improved genetic algorithm, the characteristic band of the target band is selected to obtain the effective band; Monthly features are introduced, and the chlorophyll a concentration data are selected using binary numbers 0 and 1 to obtain effective chlorophyll a concentration data. Based on the effective chlorophyll a concentration data and effective bands, the XGBoost model is constructed, and the parameters of the XGBoost model are optimized through the dung beetle optimization algorithm to obtain the optimal parameters for training the XGBoost model; The effective band data of the water area to be tested was input into the optimal parameter training XGBoost model for inversion to obtain the predicted results of chlorophyll a concentration in the water area to be tested.

[0006] Preprocessing the remote sensing image of the target water area to obtain the target band includes the following steps: After performing atmospheric correction on the remote sensing image of the target water area, a corrected remote sensing image is obtained; The corrected remote sensing image is subjected to target band extraction to obtain a remote sensing image of the target band; MNDWI is used to extract water areas from the remote sensing images of the target band to obtain the water body remote sensing images of the target band; Then, the nearest collar interpolation resampling method is used to map the image data of the water body remote sensing image of the target band, obtain the band information table data, and then obtain the target band.

[0007] The method of optimizing the parameters of the XGBoost model by using the Dung Beetle Optimization Algorithm (DBO) to obtain the optimal parameters for training the XGBoost model includes the following steps: Initialize the dung beetle algorithm population and parameters; Calculate the fitness of the dung beetle population; The dung beetle population position and fitness are updated through rolling, dancing, reproduction, foraging and stealing strategies; Update the position and fitness of the best dung beetle individual; Determine whether the set stop condition is met. If so, output the position of the best dung beetle individual, that is, the best parameters of the XGBoost model. Otherwise, update the position and fitness of the best dung beetle individual until the set stop condition is met, and output the position of the best dung beetle individual, that is, the best parameters of the XGBoost model. Get the best parameters to train the XGBoost model.

[0008] The method of selecting a characteristic band for a target band based on an improved genetic algorithm comprises the following steps: The information of 12 characteristic bands extracted from remote sensing images was input into the feature selection method, and combined with the monthly characteristic data and the measured chlorophyll a concentration data, it was input into the DBO-XGBoost model. According to the generated different binary sequences, composed of 0 and 1, corresponding to the non-selection and selection of the 12 characteristic bands, the chlorophyll a concentration of Wuliangsu Lake was predicted. According to the inversion results, the binary sequence corresponding to the best accuracy was retained, that is, the best characteristic band combination selected.

[0009] A system for remote sensing inversion of chlorophyll a concentration in waters, comprising: A data acquisition module is used to collect chlorophyll a concentration data of the target water area and simultaneously obtain a pre-processed remote sensing image of the target water area; The data processing module is used to introduce monthly features, use binary numbers 0 and 1 to represent, select chlorophyll a concentration data, obtain effective chlorophyll a concentration data, perform atmospheric correction on remote sensing images and extract remote sensing images of target bands, and select characteristic bands of target bands based on improved genetic algorithms to obtain effective bands; The inversion model building module is used to take the effective chlorophyll a concentration data and the effective band as input to build the XGBoost model, optimize the parameters of the XGBoost model through the dung beetle optimization algorithm DBO, and obtain the optimal parameters to train the XGBoost model; The inversion prediction module is used to input the data of the water area to be tested into the optimal parameter training XGBoost model for inversion, and obtain the predicted results of chlorophyll a concentration in the water area to be tested.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: using Sentinel-2 remote sensing image data, taking month information as one of the input features, using the sensitive bands for screening chlorophyll a concentration, and combining the XGBoost model to invert the chlorophyll a concentration. Among them, by combining remote sensing images and chlorophyll a concentration data, and using the XGBoost model for training, the chlorophyll a concentration of the target waters can be predicted more accurately, and the accuracy of the chlorophyll a concentration prediction is improved. By selecting the bands of remote sensing images using an improved genetic algorithm, the bands that contribute most to the prediction of chlorophyll a concentration can be effectively screened out. Data redundancy is reduced, and the efficiency and prediction performance of the model are improved. By optimizing the parameters of the XGBoost model using the dung beetle optimization algorithm, the optimal model parameter settings can be found. The generalization ability and prediction stability of the model are improved. By introducing the monthly feature, the model can consider the impact of time changes on chlorophyll a concentration. The adaptability of the model to seasonal changes is enhanced.

[0011] The results show that compared with other feature selection methods, the number of selected bands is the least, and the XGBoost model has the highest inversion accuracy, which can accurately invert the chlorophyll a concentration and improve the accuracy and efficiency of chlorophyll a concentration prediction. For water quality issues in water bodies, the proposed XGBoost model can timely and accurately monitor and warn of water quality changes, provide assistance for water quality management, and make water quality testing data public, enhance public awareness of water protection, and provide scientific data support for governments and environmental protection agencies to formulate effective agricultural and water resources management policies.

[0012] In summary, the XGBoost model not only achieves high-precision inversion of chlorophyll a concentration, but also provides an important tool for aquatic ecological environment protection, contributing to the sustainable management and healthy development of lakes. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0014] Figure 1 It is a schematic diagram of the process structure of the present invention; Figure 2 is a flow chart of DBO-XGBoost chlorophyll a concentration inversion in the present invention; Figure 3 is a bar graph of the chlorophyll a concentration inversion results of each model in the embodiment of the present invention; Figure 4 is a scatter plot of the chlorophyll a concentration inversion results of each model in the embodiment of the present invention; Figure 5 is a scatter plot of the inversion results of optimizing the XGBoost model using different intelligent optimization algorithms in an embodiment of the present invention; Figure 6 is a schematic diagram of band information selected by different feature selection algorithms in an embodiment of the present invention; Figure 7 It is the inversion result of the DBO-XGBoost model using different feature selection algorithms in the embodiment of the present invention; Figure 8 is a single-point crossover and single-point mutation operation diagram in an embodiment of the present invention; Fig. 9 In the embodiment of the present invention, the individual generates a reverse individual through reverse learning; Fig.10 is a flow chart of generating a new population through reverse learning in an embodiment of the present invention; Fig.11 is a schematic diagram of an AND operation and an OR operation in an embodiment of the present invention; Fig.12is a box plot of chlorophyll a concentration in Wuliangsuhai Lake in an embodiment of the present invention; Fig.13 is a distribution map of the geographical location and sampling points of Wuliangsuhai Lake in an embodiment of the present invention; Fig.14 It is the spatiotemporal distribution diagram of the chlorophyll a concentration of Wuliangsu Lake inverted in the embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0016] With the widespread application of remote sensing technology in the field of chlorophyll a concentration prediction in water bodies, a problem is faced: remote sensing spectral data are usually of high dimension, and there is collinearity between characteristic bands. The input of all characteristic bands maximizes the complexity of the model, and the prediction accuracy may be affected. Therefore, when establishing a chlorophyll a concentration inversion model, selecting characteristic bands that are sensitive to chlorophyll a concentration is an important prerequisite for streamlining the model, improving prediction accuracy, and enhancing generalization ability. However, it is difficult to express the relationship between remote sensing spectral features and water body elements with a simple linear relationship, so its adaptability is poor. The empirical method and analytical method are based on the combination of remote sensing data and parameters related to the inherent optical characteristics of water bodies, but they are highly dependent on the quality of spectral data and are easily affected by the concentration of other substances in the water. The inversion effect is poor when facing complex water bodies, resulting in low accuracy of prediction results.

[0017] In order to overcome the above-mentioned shortcomings of the prior art, the main purpose of the present invention is to provide a remote sensing inversion method for chlorophyll a concentration in water areas. Figure 1 , the specific steps include: Acquire the chlorophyll a concentration data and remote sensing images of the target waters at the same time; pre-process the remote sensing images of the target waters to obtain the target band; Based on the improved genetic algorithm, the characteristic band of the target band is selected to obtain the effective band; Monthly features are introduced, and the chlorophyll a concentration data are selected using binary numbers 0 and 1 to obtain effective chlorophyll a concentration data. The effective chlorophyll a concentration data and effective bands are used as input to build the XGBoost model. The parameters of the XGBoost model are optimized through the Dung Beetle Optimization Algorithm DBO to obtain the optimal parameters for training the XGBoost model. The effective band data of the water area to be tested was input into the optimal parameter training XGBoost model for inversion to obtain the predicted results of chlorophyll a concentration in the water area to be tested.

[0018] Example 1 Combination Figures 1 to 14, choose Wuliangsuhai. Wuliangsuhai (N40°47′~41°03′, E108°43′~108°57′) is located in the western part of Inner Mongolia Autonomous Region and the top of the "Jizi Bend" of the Yellow River in Urad Front Banner, Bayannur City, Inner Mongolia. It is one of the eight largest freshwater lakes in China and the largest wetland on the same latitude on the earth. It has a total area of ​​293 square kilometers, an average water depth of about 1.5 meters, and a water storage capacity of about 3-4×108 cubic meters. At the same time, Wuliangsuhai is an important habitat for a variety of organisms, with rich phytoplankton, birds, fish, zooplankton and amphibian resources. Before 2012, due to the unreasonable discharge of industrial wastewater, farmland drainage and urban domestic wastewater, the water quality of Wuliangsuhai was Class IV or even Class V, and the water body was highly eutrophic. After 2012, the ecological water replenishment of Wuliangsuhai Lake increased significantly, the heavy polluting enterprises were relocated and closed, and the non-point source pollution of farmland was reduced, which led to a certain improvement in the water quality of Wuliangsuhai Lake. However, it is still necessary to continuously monitor its water quality to ensure the continuous optimization of the ecological environment of Wuliangsuhai Lake. Taking Wuliangsuhai Lake as the target water area, the geographical location and sampling points of Wuliangsuhai Lake are as follows: Fig.13 shown.

[0019] For the chlorophyll a concentration data, the research data was provided by the national scientific research team of "River, Lake and Wetland Water Environment Protection and Restoration" of Inner Mongolia Agricultural University. The team has long been committed to the scientific research of Wuliangsu Lake, so the data source is authentic and reliable. The research data were collected at 19 sampling points in Wuliangsu Lake. However, since Wuliangsu Lake begins to freeze from late October to early November every year and begins to thaw in March of the following year, the freezing period lasts for five months, so the sampling time is concentrated from June to September. The sampling time is fixed in the middle and late of each month. Sampling is carried out by boat at the sampling point, and the sampling depth is 0.5 meters vertically downward from the water surface. A total of 91 data were sampled from 2015 to 2018, and some of the measured data were classified according to the internationally recognized degree of lake eutrophication as shown in Table 1. The internationally recognized degree of lake eutrophication is classified according to the chlorophyll a concentration content, and the lake nutrient control standard is divided into six levels, as shown in Table 2. This application divides 91 data into training set data and test set data in a ratio of 7:3, uses 63 training set data to train the model, and after the training is completed, uses 28 test set data to test and evaluate the model effect.

[0020] Table 1 Some measured data of chlorophyll a concentration

[0021] Table 2 Lake eutrophication degree classification table

[0022] The remote sensing satellite data comes from the European Space Agency's Sentinel-2A and Sentinel-2B satellites, which are mainly used to detect various information such as agriculture, forestry planting, lake water and offshore water pollution. The Sentinel-2 series satellites are all equipped with the same multispectral imager MSI, which can capture images in 13 different bands, covering visible light, near infrared and short-wave infrared ranges, with band resolutions of 10m, 20m and 60m respectively. The specific band information is shown in Table 3. At the same time, the Sentinel-2 series satellites can achieve systematic photography of land, coastal waters, and the Mediterranean from 56 degrees south latitude to 84 degrees north latitude. It covers the entire world every 5 days with an observation width of 290 kilometers, which enables it to provide frequent and high-resolution images of the earth's surface.

[0023] Table 3 Sentinel-2 satellite band information

[0024] The remote sensing satellite data is selected to be synchronized or quasi-synchronized with the sampling date of the measured chlorophyll a data, and the remote sensing satellite data of the study area is selected to be synchronized or quasi-synchronized with the sampling date of the measured chlorophyll a data, and the cloud coverage area over the study area is less than 20%. The remote sensing images of this embodiment are downloaded from the Copernicus Data Center of the European Space Agency.

[0025] For data processing and remote sensing image preprocessing, the remote sensing image data downloaded in this embodiment is of Level-1C. The atmospheric correction of Level-1C remote sensing images is performed through the Sentinel-2 image processing software Sen2Cor provided by the European Space Agency to generate Level-2A remote sensing image data. In addition, since band 10 is a cumulus cloud band, Sen2Cor does not process this band, that is, the final result is 12 bands of data except band 10. The processed Level-2A remote sensing image data is extracted using MNDWI for water area extraction, and then resampled using the nearest neighbor interpolation method.

[0026] Regarding monthly characteristics, in the global water eutrophication monitoring, some scholars have studied the influence of different seasonal data training models on the inversion accuracy based on the measured chlorophyll a concentration of 38 major lakes in the Yangtze River Basin and Sentinel-2 remote sensing data. The results show that the sampling time significantly affects the performance of the chlorophyll a concentration inversion model of the lake. At the same time, many domestic scholars have also found that the eutrophication degree of Wuliangsuhai Lake has obvious seasonal characteristics in their research on the water quality of Wuliangsuhai Lake. Based on the measured chlorophyll a concentration data of Wuliangsuhai Lake, a box plot is drawn as follows Fig.12As shown, it can be seen that the chlorophyll a concentration of Wuliangsu Lake varies between June, July, August, and September. The chlorophyll a concentration from July to August varies greatly, with the average value differing by as much as three times. Therefore, this embodiment introduces the monthly feature, and uses the month information of the data, such as June, July, August, and September, as features, represented by binary numbers 0 and 1. When 1, it means that the data was collected in that month, otherwise it means that it was not collected in that month. The final data feature structure is shown in Table 4.

[0027] Table 4 Experimental data feature structure

[0028] The task of feature selection is to select a sub-feature set that meets the conditions from a given feature set. In a problem with a feature scale of N, each feature has two states: selected with a binary number 1 and not selected with a binary number 0, with a total of 2N combinations. Using an exhaustive method consumes a lot of time and has high computational costs. The genetic algorithm GA searches for the optimal solution in the solution space by imitating the selection, crossover, and mutation operations in biological evolution in nature. It is a highly flexible and robust optimization method. Therefore, using a genetic algorithm to select feature bands can efficiently filter out redundant band information, reduce model complexity, and improve model accuracy.

[0029] The traditional genetic algorithm GA is mainly divided into the following steps: (1) Initialize the population: Use a binary number (0, 1) to indicate whether the feature is selected. 0 means that the feature is not selected, and 1 means that the feature is selected.

[0030] (2) Selection: Calculate the fitness of all individuals in the population and use the roulette wheel selection algorithm to select individuals to enter the next generation.

[0031] (3) Crossover: This is the global search phase in the entire genetic algorithm. Two parent individuals are randomly selected by probability, and the chromosomes of the two individuals are crossed. This embodiment uses the single-point crossover method, and the operation is as follows: Figure 8 shown.

[0032] (4) Mutation: Use single-point mutation to randomly generate a mutation point, changing 0 to 1 and 1 to 0. Figure 8 shown.

[0033] To improve the genetic algorithm (1) Introducing population initialization for reverse learning Opposition-based-learning (Obl) was proposed by Tizhoosh in 2005 and has been successfully applied in swarm intelligence optimization algorithms such as GA, ant colony algorithm and differential evolution algorithm. The reverse solution is calculated from the current candidate solution, and the candidate solution and the reverse solution are evaluated, and finally the better solution is selected as the final solution. The reverse solution is defined as: Assume that the candidate solution is a point in n-dimensional space , and its reverse solution is , Calculate as formula (1): (1) for A random number uniformly distributed on for The upper limit of for After each iteration generates the initial population, the reverse solution is obtained through the formula: Expressed as a fitness evaluation function, for the minimization problem, if , then select As the final solution, otherwise choose .

[0034] Generate reverse individuals through reverse learning strategy, such as Fig. 9 shown.

[0035] Assuming the population size is N, calculate the reverse individual of each initially randomly generated individual, and select the individuals and reverse individuals with better fitness to form a new population, such as Fig.10 By generating a new population through the reverse learning strategy, the search range of the initial population is enhanced and the fitness of the initial population is improved, thereby accelerating the convergence of the algorithm and saving computing time.

[0036] (2) Adaptive crossover mutation probability: The crossover probability and mutation probability of traditional genetic algorithms are fixed, and each individual has the same crossover and mutation probability, which affects the algorithm's search ability and convergence speed. Therefore, the crossover and mutation probabilities are improved so that they change according to individual fitness and population fitness, enhancing the algorithm's search ability and accelerating convergence.

[0037] In the AND or crossover algorithm, the characteristics of the parent generation are retained to the greatest extent. Therefore, individuals with better fitness need to have a higher crossover probability so that their characteristics can be inherited to the next generation; while individuals with poor fitness should have a lower crossover probability to reduce the inheritance of their characteristics to the next generation. In order to solve the minimization problem in this application, the adaptive crossover probability is defined as follows (2): (2) in, , are set to 0.9 and 0.4, respectively, corresponding to the highest crossover probability and the lowest crossover probability, To select the better fitness of the two individuals. represents the average fitness of the contemporary population, Represents the optimal fitness in the contemporary population.

[0038] The mutation operation prevents the population from falling into the local optimum by randomly changing the genes of individuals. For individuals with better fitness, it is hoped that their mutation probability is low to protect the good gene structure, while for individuals with poor fitness, it is hoped that their mutation probability is high to explore new solution space. Therefore, the adaptive mutation probability for solving the minimization problem is defined as follows: (3) in, , are set to 0.1 and 0.01, corresponding to the maximum mutation probability and the minimum mutation probability, respectively. is the fitness of the selected mutant individual. represents the average fitness of the contemporary population, Represents the optimal fitness in the contemporary population.

[0039] c. AND or crossover algorithm A crossover algorithm combining "AND-OR" logic is proposed, which is mainly divided into the following four steps: (0) According to the adaptive probability, two individuals are selected as parent 1 and parent 2; Perform a bitwise AND operation on parent 1 and parent 2 as shown in equation (4) to generate Child 1. Child 1 retains the common features of parent 1 and parent 2. The bitwise AND operation is as follows: Fig.11 shown.

[0040] (4) (1) Perform a bitwise OR operation on parent 1 and parent 2 as shown in equation (5) to generate Child 2. Child 2 retains all the features of parent 1 and parent 2. The bitwise OR operation is as follows: Fig.11 shown.

[0041] (5) Child1 generated by the bitwise AND operation only retains the features shared by the parent generation, i.e., retains the least features. Child2 generated by the bitwise OR operation retains all the features of the parent generation, i.e., retains the most features. Therefore, in order to ensure the diversity of the offspring population, Child1 and Child2 are compared bit by bit. When a certain bit is different, 0 or 1 is randomly selected for that position of Child1 until Child1 is traversed.

[0042] The Dung Beetle Optimizer Algorithm (DBO) is a swarm intelligence algorithm proposed by XUE in 2023. It updates its position by simulating the rolling, dancing, reproduction, foraging, and stealing behaviors of dung beetles, while considering global exploration and local development. It has the characteristics of fast convergence speed and high accuracy, and can effectively solve complex optimization problems.

[0043] The XGBoost algorithm was proposed by Chen et al. in 2016 and is an improvement on the GBDT algorithm. By introducing the second-order derivative, XGBoost has a higher improvement in accuracy compared to the GBDT algorithm. At the same time, XGBoost adds a regularization term to the objective function to control the complexity of the model, making the learned model simpler and helping to prevent overfitting. The setting of parameters has an important impact on the model effect, but since the XGBoost algorithm has many parameters and the parameter adjustment is more complicated, DBO is used to optimize the XGBoost algorithm parameters. Table 4 shows the parameters for XGBoost tuning by DBO in this application.

[0044] Table 4 XGBoost tuning parameters

[0045] By using the DBO algorithm to tune the XGBoost model parameters, the complex problem of multi-tuning of XGBoost model parameters has been well solved. Figure 2 The process of DBO-XGBoost chlorophyll a concentration inversion model is shown. The specific steps of DBO-XGBoost chlorophyll a concentration inversion model are as follows: (2) Input remote sensing image data of Wuliangsu Lake and measured data of chlorophyll a concentration; (3) Build the XGBoost model; (4) Initialize the dung beetle algorithm population and parameters; (5) Calculate the fitness of the dung beetle population; (6) Update the position and fitness of the dung beetle population through rolling, dancing, reproduction, foraging and stealing behavior update strategies; (7) Update the position and fitness of the best dung beetle individual; (8) Determine whether the stop condition is met. If so, execute step (8); otherwise, execute step (5); (9) Output the position of the best dung beetle individual (i.e., the optimal parameters of the XGBoost model); Train the XGBoost model based on the optimal parameters obtained by the DBO algorithm; (10) The predicted results of chlorophyll a concentration in Lake Wulansuhai were obtained, and R2, RMSE, and MAE were calculated to evaluate the model performance.

[0046] This application uses mean absolute error (MAE), root mean square error (RMSE) and determination coefficient (R2) as evaluation indicators to evaluate the experimental accuracy of the model. Both MAE and RMSE are used to measure the difference between the model prediction value and the measured value. MAE is not sensitive to outliers. Due to the existence of squares, RMSE is more sensitive to large errors. The smaller the MAE and RMSE, the smaller the error and the better the model inversion effect. 2 It directly reflects the inversion effect of the model, and its range is generally from 0 to 1. 2 The closer it is to 1, the better the model inversion effect. 2 The calculation formula is (6) (7) (8): (6) (7) (8) Where n represents the number of samples, represents the measured value of chlorophyll a concentration, represents the inverted value of chlorophyll a concentration, It represents the average value of the measured values ​​of chlorophyll a concentration.

[0047] Example 2 A specific embodiment is given by applying the scheme of the present application, in which the model parameter settings have a crucial impact on the effectiveness and stability of the model. The GA and DBO algorithm parameter settings are shown in Table 5 and Table 6 respectively.

[0048] Table 5 GA and IGA algorithm parameter settings

[0049] Table 6 DBO algorithm parameter settings

[0050] In order to verify the influence of monthly characteristics on the inversion of chlorophyll a concentration in Lake Wuliangsuhai, this application established XGBoost, SVR and RF models for comparison with and without introducing monthly characteristics. The experimental results are shown in Table 7. Figure 3 The bar chart of the experimental results of the six models shows that the R² of the three models of XGBoost, SVR and RF, which introduce the month feature, is greater than 0.75, the RMSE is lower than 4.5μg / L, and the MAE is less than 3.6μg / L. Compared with the model without the introduction of the month feature, R 2 The results showed that after the introduction of the monthly feature, the R² of all models increased significantly, while the RMSE and MAE decreased, which verified that the monthly feature had a significant impact on the prediction accuracy of the model. Among all the models, the R² of the XGBoost model with the introduction of the monthly feature was 0.7916, and the RMSE and MAE were 4.1330μg / L and 3.2803μg / L, respectively, which were better than SVR and RF. Figure 4 The scatter plots of the inversion results of chlorophyll a concentration in Wuliangsu Lake by six models are shown, where the straight line is a 1:1 straight line, representing the ideal state - the inversion value is completely equal to the measured value. It can be seen that the chlorophyll a concentration inversion value of the M-XGBoost model has the smallest difference with the measured value and is closest to the 1:1 straight line, indicating that its inversion effect is the best.

[0051] In summary, the introduction of monthly features significantly improved the prediction accuracy of each model, especially the XGBoost model, which is most suitable for the inversion task of chlorophyll a concentration in Wuliangsu Lake. Therefore, in subsequent experiments, monthly features were introduced and inversion tasks were performed based on the XGBoost model.

[0052] Table 7 Chlorophyll a concentration inversion results of each model

[0053] XGBoost has many parameters and is complex to tune. Simple grid search, random search, etc. cannot find the most suitable parameter combination. Therefore, this application uses intelligent optimization algorithms to adjust XGBoost parameters, including particle swarm optimization algorithm, gray wolf optimization algorithm, and dung beetle optimization algorithm. After introducing the monthly characteristics, the accuracy of each model in inverting the chlorophyll a concentration in Wuliangsuhai is analyzed, and the results are shown in Table 8.

[0054] It can be seen that the XGBoost model optimized by the intelligent optimization algorithm has a higher R 2 All of them are greater than 0.86, showing good fitting and generalization capabilities. Secondly, the RMSE is less than 3.3 μg / L, and the MAE is less than 2.8 μg / L. The inversion accuracy of the optimized model has been greatly improved. 2 is 0.8820, RMSE and MAE are 3.1111μg / L and 2.4829μg / L, respectively. Figure 5 The scatter plots of the inversion results of the four models are shown. It can be seen that the chlorophyll a concentration prediction value of the DBO-XGBoost model is closest to the 1:1 line and has the smallest deviation from the measured value, indicating that its inversion accuracy is the best. In summary, in the task of inverting the chlorophyll a concentration of Wuliangsu Lake, using DBO to tune the XGBoost model parameters can effectively improve the inversion accuracy of the model and reduce the inversion error.

[0055] Table 8 Inversion results of XGBoost model optimized by different intelligent optimization algorithms

[0056] In order to verify the performance of the feature selection method proposed in this application, this application inverts the chlorophyll a concentration of Wulansuhai Lake based on AllBand, CARS, GA and four feature selection methods, combined with the DBO-XGBoost model. Figure 6 The experimental results are shown in Table 9. After the introduction of CARS, only 7 feature bands were selected. Compared with the DBO-XGBoost model with full-band input, the complexity of the model was reduced, and the R2 and RMSE were improved. The feature selection models of GA and GA selected 6 and 5 features respectively, which further reduced the complexity of the model and greatly improved the inversion accuracy. Among them, the DBO-XGBoost model had the best R2. 2 It was 0.9379, and the RMSE and MAE were reduced to 2.2566μg / L and 1.9019μg / L, respectively. Figure 7 The figure is a scatter plot of the inversion results of the four models. It can be seen that the chlorophyll a concentration prediction value of the DBO-XGBoost model is closest to the 1:1 line and has the smallest deviation from the measured value. The DBO-XGBoost model has the best fitting degree and the best inversion effect.

[0057] Table 9 DBO-XGBoost model inversion results of different feature selection algorithms

[0058] Through the above experimental comparison, the DBO-XGBoost model proposed in this application has the best effect in the inversion of chlorophyll a concentration in Wuliangsu Lake. Based on this model, the inversion results of chlorophyll a concentration in Wuliangsu Lake were obtained, and the distribution map of chlorophyll a concentration in Wuliangsu Lake was drawn as shown in the figure. Fig.14As shown in the figure. As can be seen from the figure, the chlorophyll a concentration in Wuliangsuhai Lake has obvious monthly characteristics. The chlorophyll a concentration in June and July is generally low, concentrated in 1-12µg / L. In August and September, the chlorophyll a concentration is relatively high, generally higher than 10µg / L, and basically concentrated in 12µg / L-30µg / L. In particular, it reached a peak in August 2017, with some areas exceeding 36µg / L. Overall, it can be clearly seen that the chlorophyll a concentration in the northern region of Wuliangsuhai Lake is generally higher than that in the southern region. In the spatiotemporal distribution map of the chlorophyll a concentration in Wuliangsuhai Lake in six time periods, the peak of the chlorophyll a concentration is located in the central region. This is because the agricultural irrigation wastewater discharge ditch is located in the middle of Wuliangsuhai Lake, and there are a large number of reeds distributed in the northern region. Pollutants gather in the northern region, resulting in a higher chlorophyll a concentration in the northern region and a higher degree of eutrophication than in the southern region. These findings are highly consistent with the research results of other scholars.

[0059] This application uses Sentinel-2 remote sensing image data, and takes the month information as one of the input features. It uses the sensitive bands of chlorophyll a concentration to screen, and combines the XGBoost model to invert the chlorophyll a concentration of Wuliangsu Lake. The results show that compared with other feature selection methods, the number of screened bands is the least, and the DBO-XGBoost model has the highest inversion accuracy, which can accurately invert the chlorophyll a concentration of Wuliangsu Lake.

[0060] In response to the water quality problems facing Wulansuhai Lake, the DBO-XGBoost model proposed in this application can timely and accurately monitor and warn of water quality changes, provide assistance for water quality management, and make public water quality test data to enhance public awareness of Wulansuhai Lake protection, and provide scientific data support for the government and environmental protection agencies to formulate effective agricultural and water resources management policies. In summary, the -DBO-XGBoost model not only achieves high-precision inversion of chlorophyll a concentration, but also provides an important tool for the ecological and environmental protection of Wulansuhai Lake, and promotes the sustainable management and healthy development of the lake.

[0061] It should be noted that, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0062] The above embodiments are merely examples of the present invention and do not limit the protection scope of the present invention. All designs that are the same or similar to the present invention fall within the protection scope of the present invention.

Claims

1. A remote sensing inversion method for chlorophyll a concentration in waters, characterized in that: The following steps are involved: Acquire the chlorophyll a concentration data and remote sensing images of the target waters at the same time; pre-process the remote sensing images of the target waters to obtain the target band; Based on the improved genetic algorithm, the characteristic band of the target band is selected to obtain the effective band; Monthly features are introduced, and the chlorophyll a concentration data are selected using binary numbers 0 and 1 to obtain effective chlorophyll a concentration data. Based on the effective chlorophyll a concentration data and effective bands, the XGBoost model is constructed, and the parameters of the XGBoost model are optimized through the dung beetle optimization algorithm to obtain the optimal parameters for training the XGBoost model; The effective band data of the water area to be tested was input into the optimal parameter training XGBoost model for inversion to obtain the predicted results of chlorophyll a concentration in the water area to be tested.

2. The remote sensing inversion method for chlorophyll a concentration in water area according to claim 1, characterized in that: The step of obtaining the target band comprises the following steps: After performing atmospheric correction on the remote sensing image of the target water area, a corrected remote sensing image is obtained; The corrected remote sensing image is subjected to target band extraction to obtain a remote sensing image of the target band; MNDWI is used to extract water areas from the remote sensing images of the target band to obtain the water body remote sensing images of the target band; The image data of the water body remote sensing image of the target band is mapped using the nearest collar interpolation resampling method to obtain the band information table data, and then the target band is obtained.

3. The method for remote sensing inversion of chlorophyll a concentration in water area according to claim 1, characterized in that: The method of obtaining the optimal parameters for training the XGBoost model comprises the following steps: Initialize the dung beetle algorithm population and parameters; Calculate the fitness of the dung beetle population; Update the position and fitness of the dung beetle population through rolling, dancing, reproduction, foraging and stealing behavior update strategies; update the position and fitness of the best dung beetle individual; Determine whether the set stop condition is met. If so, output the position of the best dung beetle individual, that is, the best parameters of the XGBoost model. Otherwise, update the position and fitness of the best dung beetle individual until the set stop condition is met, and output the position of the best dung beetle individual, that is, the best parameters of the XGBoost model. Get the best parameters to train the XGBoost model.

4. The remote sensing inversion method for chlorophyll a concentration in water area according to claim 1, characterized in that: The method of selecting a characteristic band for a target band based on an improved genetic algorithm comprises the following steps: Set the initial population, each population represents a band combination; Evaluate the fitness of each band combination and use the training set to evaluate the classification accuracy of the classifier on the selected band combination; Select individuals whose fitness is higher than the set threshold range as the parents of the next generation of individuals; Apply crossover and mutation operations to generate new individuals; The population is updated to generate a new generation, and the evaluation, selection, crossover and mutation steps are repeated until the stop condition is reached to obtain remote sensing images of valid bands.

5. A system for remote sensing inversion of chlorophyll a concentration in waters, characterized in that: include: A data acquisition module is used to collect chlorophyll a concentration data of the target water area and simultaneously obtain a pre-processed remote sensing image of the target water area; The data processing module is used to introduce monthly features, use binary numbers 0 and 1 to represent, select chlorophyll a concentration data, obtain effective chlorophyll a concentration data, perform atmospheric correction on remote sensing images and extract remote sensing images of target bands, and select characteristic bands of target bands based on improved genetic algorithms to obtain effective bands; The inversion model building module is used to take the effective chlorophyll a concentration data and the effective band as input to build the XGBoost model, optimize the parameters of the XGBoost model through the dung beetle optimization algorithm, and obtain the optimal parameters to train the XGBoost model; The inversion prediction module is used to input the data of the water area to be tested into the optimal parameter training XGBoost model for inversion, and obtain the predicted results of chlorophyll a concentration in the water area to be tested.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of claims 1 to 4 is implemented.

Images