| Satellite remote sensing technology plays an irreplaceable role in estuarine and coastal zone monitoring.Due to its high spectral resolution,hyperspectral has significant advantages in the identification of ground object information and inversion of surface parameters,and has become one of the development trends of satellite remote sensing technology.Gaofan-5 satellite is one of the hyperspectral satellites launched by China in recent years,but the application of the satellite image to the related technology research needs to be further carried out.By take the Yangtze River Estuary as the research area,based on Hyperspectral image of GF-5 and combined with the satellite and ground synchronous survey,this research studied the extraction technology of vegetation distribution in the tidal flat of the Yangtze River Estuary and the inversion technology of typical water quality parameters.A variety of remote sensing models was developed to verify the innovative achievements of China’s high-resolution satellite,then the differences of each model were compared,the reasons of the differences were analyzed.The main conclusions of this paper are as follows:1)Research on influence extraction based on Gaofen-5 satellite remote sensing image.Aiming at the spatial distribution of plant communities in tidal flats of the Yangtze River Estuary,the high-precision extraction technology based on traditional supervised classification was studied,and accurate extraction of typical tidal flat plant communities in the Yangtze River Estuary was achieved(with an accuracy of 95.4%).This result confirms the superiority of Gaofen-5.At the same time,a set of efficient interpretation models based on machine learning algorithm has been developed,which also achieves a high interpretation accuracy(with an accuracy of 92.7%).2)Research on the water quality of inversion based on Gaofen-5 satellite remote sensing image.The numerical inversion of several water quality parameters were realized by study the inversion of water quality parameters in the Yangtze River Estuary.For most water quality parameters,the optimal inversion model generally has a very high level in all parameters,with the lowest R~2 of 0.856,and the average relative error of most parameter models is less than 10%,while the average relative error of the optimal model for only one water quality parameter is greater than 15%.3)By compare band combination inversion and random forest model,it is known that when the environmental conditions are unknown,random forest algorithm can output a better model stably while the output model of band combination method needs to be further verified in practical operation.4)In the band combination method,by compare linear models and exponential models,it is known that the exponential model always has good performance in the vast majority of cases.But the application potential of the linear model should not be ignored when the numerical distribution interval meets the requirements of the linear model,which means the ratio of the minimum value to the maximum value of the parameter is large enough.One the other hand,by comparing the combinations of various bands,it can be concluded that the semi-analytical algorithm has obvious advantages over the ordinary empirical algorithm.5)In the random forest algorithm,by trying to remove non-important bands,it is known that there is no need for special screening of the bands unless there is a special need.By comparing the performance of the model before and after the linear correction,it can be seen that the relative error and root mean square error of the model can be reduced by the linear correction under certain conditions,especially when the intermediate parameters are corrected by the linear correction when the exponential model is used. |
