Research On Land Use Changes Prediction Model Based On Adaptive Variable Filter

With the development of the economy and the growth of the population,the scale of land development and construction and the land use situation is changing year by year.Accurate prediction of land use in the future can provide scientific and rational land use planning for the region.The land use in the region will be more reasonable.The research area selected in this paper is the Xiangfang and the Pingfang area in Harbin.The three-year land use situation is analyzed for dynamic change,and multiple models are constructed to predict land use in these two areas in 2020.The main research options are as follows:Landsat remote sensing images from Harbin Xiangfang and Pingfang area in 2005,2010 and 2015 are selected as the basic data,and ENVI5.1 and ArcGIS10.2 is used to interpret the data into a grid format with a resolution of 30-100 meters,and divide the Research area into seven categories: cultivated land,forest land,grassland,urban land,rural residential land,other construction land and unused land.The raster data is clipped as the training data according to the range of the rectangular frame outside the range,and the data of the area to be predicted and its surrounding area is used as the input data.After the model prediction result is obtained,the vector file of the area is used to tailor again,which improves the efficiency of data usage and eliminates the impact of data missing on the accuracy of predicted boundary regions.Traditional methods generally use CA_Markov,ANN,and CA_ANN models for prediction.There are problems such as long training time,insufficient prediction accuracy,and lack of persuasiveness.In view of the above problems,an adaptive variable filter network model is established to create multiple types of data sets to train multiple nerves with different parameters for the number of land categories in a specific size region.This model can not only successfully predict the future land change,but also avoid the offset of data weights on the network when training a single network.The experimental results show that compared with the traditional CA_Markov model and CA_ANN model,the overall predicting accuracy of the new method is improved by 4.46%,the predicting accuracy of various land types is improved by 2.93%-13.87%,and the training time of the model is also reduced by 66.39%.Compared with other models,the variable filter model can not only predict the future land use situation more accurately,but also take less time.In the experiment,we applied the new model to the prediction of land change in the Pingfang area.The accuracy of the experimental test was close to that of Xiangfang area,indicating that the model has good generalization performance.