Remote Sensing Retrieval Of Chlorophyll-a In Water Based On Spatio-temporal Spectrum Fusion And Depth Learning

Remote sensing technology provides a convenient way for inland water monitoring.The combination of traditional on-site sampling and remote sensing geographic information technology can quickly and widely monitor water quality.However,due to the quality of remote sensing images,the current water quality inversion products are difficult to consider both high temporal resolution and high spatial resolution.Because the remote sensing water quality inversion process is more complex,and the traditional empirical model and physical model have their own advantages and disadvantages,it is necessary to select an appropriate inversion model to adapt to different remote sensing data.Taking Baiyangdian water area as the research area,combined with water quality sampling points and multi-source remote sensing images,this paper studies the inversion method of chlorophyll-a in water body based on spatio-temporal spectrum fusion and deep learning.The spatiotemporal spectrum fusion data is generated by spatial spectrum fusion and spatiotemporal fusion algorithm,and then the chlorophylla concentration of Baiyangdian Lake is inversed combined with depth learning to generate time series.The main contents of the article include:(1)Combining GS(Gram-Schmidt)space spectrum fusion and STARFM(Spatial and Temporal Adaptive Reflectance Fusion Model)space-time fusion,a better spacetime spectrum fusion image can be obtained.In the study of spatial spectrum fusion of GF-2,it is found that the spatial spectrum fusion method based on GS fusion can obtain better results in water.The correlation between the blue,green,red and near-infrared bands of water after spatial spectrum fusion and the original multispectral bands is0.813,0.865,0.7984 and 0.953 respectively;In the space-time fusion research of GF-2 and Sentinel-2,the space-time fusion method based on STARFM fusion can obtain better results in the water body.The correlation between the blue,green,red and nearinfrared bands of the water body based on space-time fusion and the original multispectral bands can reach 0.873,0.882,0.895 and 0.944 respectively.After spatiotemporal spectrum fusion,the temporal resolution of the image reaches 5 days and the spatial resolution reaches 1 meter,which still retains the spectral characteristics of the water body.(2)After data expansion and parameter iteration,CNN(Convolutional Neural Networks)model can obtain good inversion effect on spatiotemporal spectrum fusion image.In the study,the average correlation of BP neural network is 0.565 and that of CNN neural network is 0.793.The number of convolution cores is iterated,and the 4-(236-452-996)-1-1 convolution neural network model with Sentinel-2 as the remote sensing data source is determined,with a correlation of 0.817,and the 4-(228-332-932)-1-1 convolution neural network model with GF-2 like(1m)as the remote sensing data source,with a correlation of 0.825.Finally,the remote sensing data on August 21,2019 is used to apply the inversion model.It is found that the water quality results of spatio-temporal spectrum fusion images can maintain the overall water quality distribution characteristics and provide more local spatial details.(3)The generated chlorophyll-a inversion product has high temporal and spatial resolution,which provides scientific monitoring data for the dynamic observation of cyanobacteria in Baiyangdian Lake.The retrieval of chlorophyll-a from Sentinel-2images from July to September 2016-2021 shows that there are three different temporal and spatial variation characteristics in the lake area.Based on the Sentinel-2 images from July to September 2016-2021,the spatiotemporal spectrum fusion data set was generated,and the chlorophyll-a concentration was inversed by CNN model,and the chlorophyll-a inversion product with high spatiotemporal resolution was generated.It is found that when the spatial difference is large enough,the inversion results of the fused image and the original image are highly correlated.The concentration of chlorophyll-a in Baiyangdian area is low in 2020,and the water pollution control of Baiyangdian has achieved remarkable results.