Inversion Of Chlorophyll-a Concentration In Wuliangsuhai Lake Based On BP Neural Network And Sentinel-2 Remote Sensing Image

Wetland ecological environment construction is a key link in the construction of my country’s ecological security system,and also an important basis for achieving sustainable economic and social development.Wuliangsuhai Lake Wetland is one of the most representative wetlands in arid and semi-arid regions in the world,and it is also an important breeding ground for the East Asia-Australasian Flyway and the Central Asia Flyway.From the 1990s to the present,the over-exploitation of Wuliangsuhai Lake wetland has caused serious damage to the ecological environment of Wuliangsuhai Lake,at the same time,the ecological water supply of Wuliangsuhai Lake decreased,the ecological function of Wuliangsuhai wetland is seriously degraded,the lake water body is seriously eutrophicated,and the process of swamping is accelerated.Scientific detection and development trend prediction of the ecological environment status of Wuliangsuhai Lake wetland,it has become one of the key issues that must be solved in wetland ecological environment governance.Chlorophyll-a can reflect the degree of eutrophication and pollution of lake water quality,it is an important indicator in water quality testing.BP neural network is an effective method for solving nonlinear problems and is often used in forecasting.In this thesis,based on the BP neural network,the inversion of chlorophyll-a concentration in Wuliangsuhai Lake has been studied,the monitoring of chlorophyll-a concentration in Wuliangsuhai Lake wetland was realized by analyzing remote sensing image data.The main research contents are as follows:(1)Obtained the measured data of chlorophyll-a concentration at the sampling point of Wuliangsuhai Lake and the remote sensing image data of Sentinel-2 during the same period,after preprocessing the remote sensing image data,taking the reflectivity of all bands of the remote sensing image as the input value,based on BP neural network and remote sensing images,an inversion model of chlorophyll-a concentration in Wuliangsuhai Lake was initially constructed.Use the test data to test the trained model,the results show that the coefficient of determination(R~2)of the inversion model is 0.22and the root mean square error(RMSE)is 9.56,it shows that it is feasible to use BP neural network to invert the concentration of chlorophyll-a in Wuliangsuhai Lake,however,the inversion accuracy of the initially constructed BP neural network inversion model is poor.(2)In-depth analysis of the inversion effect of the initially constructed BP neural network,the input value of the neural network is optimized by the method of correlating the reflectance of the remote sensing image band and adding the time dimension to the input data,under the premise that other conditions remain unchanged,rebuild the BP neural network inversion model.The experimental results show that the model inversion accuracy has been greatly improved after optimizing the input value,the coefficient of determination(R~2)of the inversion model is 0.61 and the root mean square error(RMSE)is 5.78。(3)BP neural network has the defects of slow convergence speed and easy to fall into local optimum,in this thesis,particle swarm algorithm combined with back propagation is used to optimize the BP neural network,while the particle swarm algorithm performs extensive search of the solution space,the back-propagation algorithm is used to replace the individual optimal and group optimal in the particle swarm.Experimental results show,the inversion accuracy of the BP neural network optimized by particle swarm optimization is further improved,the coefficient of determination(R~2)of the PSO-BP inversion model is 0.82,and the root mean square error(RMSE)is 3.77,at the same time,the BP neural network inversion model optimized by particle swarm optimization requires fewer training times to converge.To sum up,the BP neural network inversion model effectively overcomes the limitation that the traditional empirical model cannot deal with nonlinear problems,and the inversion accuracy of the model is higher after using the particle swarm optimization algorithm to optimize the BP neural network.The research shows that the use of remote sensing images combined with BP neural network to invert the concentration of chlorophyll-a in Wuliangsuhai Lake has good feasibility and broad application prospects.