Research On Database Construction And Automatic Identification Method Of Land Cover Image Based On Machine Learning

With the development of big data technology,many platforms can store and provide massive surface images.Flickr is one of them.Global surface coverage and utilization of maps has always been the basic platform for many research contents,but the surface is changing every day,which leads to the low timeliness of land cover utilization maps.At present,the commonly used methods for obtaining surface coverage maps are extracted by remote sensing image analysis.On the one hand,the accuracy of extracting content is related to the resolution and analysis technology of remote sensing images.On the other hand,obtaining global remote sensing image analysis and extracting surface coverage maps is also a huge work of engineering.Therefore,how to obtain accurate,low-cost and time-sensitive surface coverage and utilization maps is the hotspot and difficulty of current research.With the advancement of artificial intelligence,there have been many successful cases of deep learning in the field of computer vision.Applying deep learning to geography and environment research is a relatively innovative work.Big data provides massive surface image data for deep learning,but the screening and annotation of images is also difficult.At present,there are related land cover image datasets,but they are not classified according to a classification system.It is very difficult and innovative to create image datasets classified according to the surface coverage classification system.The data determines the upper limit of the model,so the accuracy of the existing land cover classification model needs to be improved.Studying how to improve the accuracy of the model is also an important research direction.Therefore,this paper obtains massive data sets based on existing network technologies,creates image data warehouses according to LCCS classification system,and studies the application of deep learning in surface coverage classification based on data warehouse.This paper first outlines the importance of land cover using maps in various fields,and the problems encountered in extracting land cover maps.It presents the existing surface coverage classification system,the development history of deep learning and the status quo of deep learning in the automatic identification of surface coverage types.It also analyzes the shortcomings of deep learning in the field of identifying surface coverage types.Secondly,on the basis of studying the construction of surface cover datasets,the paper puts forward the classification of datasets according to the LCCS classification system and constructs an index that can be autonomously based on geospatial scope and keywords.A data warehouse for the corresponding image dataset.This work provides a powerful data foundation for identifying areas of surface coverage.Finally,on the basis of the constructed data warehouse,the European region was selected as the research area to study the training of the land cover classification model in Europe,and the accuracy of the annotation data was verified.By adjusting the Dropout and the initial learning rate,the Adam and RMSProp adaptive learning rate algorithm is used to select the model algorithm with the best network recognition performance.The experimental results show that the initial learning rate of the network model is set to 0.005,the dropout is set to 0.5,and the optimization algorithm will have better performance when using the Adam algorithm.Among them,the accuracy rate is Top1 59% and Top3 83%.The model can achieve good recognition results,provide technical support for surface coverage type identification applications,and provide a new method for global surface coverage using map extraction.And the global surface coverage image data warehouse based on LCCS classification system constructed in this paper provides data support for more research content.