Remote Sensing Image Desertification Classification And Change Monitoring Based On Optimized BP Neural Network

The sixth national monitoring of desertification will be fully carried out in 2019,Accurate judgment and quantitative analysis are made on the change of desertification land in real time Not only can save manpower and material resources,but also achieve point-to-point precise governance,effectively save social resources.Lay a good foundation for the scientific management of desertified land.Do a good job in monitoring desertification land changes is an important move about protecting the ecosystem,maintain ecosystem stability,promote desertification control,Diversified ecological compensation mechanism.There are certain problems in the classification and change monitoring of desertified land by using remote sensing technology or BP neural network.Traditional remote sensing classification methods rely too much on image quality and are not highly automated.BP neural network has slow convergence and unstable memory learning during training,leading to remote sensing image classification accuracy is not high.In this paper,a BP neural network remote sensing image classification method based on GA and PSO optimization is proposed and applied to remote sensing image desertification classification and change monitoring.The research work of this paper mainly includes:1)The realization process and advantages and disadvantages of BP neural network,GA,PSO and Alexnet convolutional neural network model are studied.Based on this,a BP neural network remote sensing image classification method based on GA and PSO optimization is proposed,apply it to remote sensing image classification.The experimental results show that the GA-PSO-BP remote sensing image classification method has a fast convergence rate and stable learning,which can improve the classification accuracy of remote sensing images.2)Taking the remote sensing image of some areas of Yanshan Town,Guilin,Guangxi as the training area.,divide the ground objects in the training area into 6 categories.According to certain principles and different scales,500 multi-scale samples were collected.Image preprocessing work is performed on the training area and the test area image to reduce the classification error caused by the image itself.3)Remote sensing image of training area based on training samples and pre-processing,get on BP,GA-BP,PSO-BP,GA-PSO-BP,CNN-Alexnet neural network remote sensing image classification experiments.The accuracy of the confusion matrix analysis is carried out on the experimental results.The experimental results show that the GA-PSO-BP neural network model is simple and easy to operate,the classification efficiency is high precision,and the simulation error for the sample data set is 10.0104.4)Classification of images in the test area in 2005,2010 and 2016 based on GA-PSO-BP neural network remote sensing image classification method.The difference method was used to monitor the change of classification results.Finally,the reasons for the changes are analyzed and summarized in the change monitoring results,and reasonable Suggestions for governance measures are put forward.The GA-PSO-BP neural network remote sensing image classification method proposed in this paper can effectively improve the slow convergence speed and unstable memory learning of BP neural network,and the structure is simple and easy to operate.There is no specific requirement for the size of the sample,which can be better.Improve the accuracy of remote sensing image classification and change monitoring.