Research On Rocky Desertification Information Extraction Based On Improved BP Neural Networks

In order to improve the accuracy of remote sensing data of rocky desertification area and the slope of rocky desertification area,this paper mainly extracts the high-resolution remote sensing data of rocky desertification area,and improves the accuracy of remote sensing data Based on the BP neural network model,the initial BP neural network model is optimized by gray wolf optimization algorithm(GWO),whale optimization algorithm(WOA),particle swarm optimization algorithm(PSO)and differential evolution algorithm(DE),and the sample features are classified by the BP neural network before and after optimization.The following conclusions are drawn:(1)In the karst area of Puding County in 2021,the areas of non-rocky desertification(NRD),potential rocky desertification(PRD),mild rocky desertification(MRD),moderate rocky desertification(MORD)and severe rocky desertification(SRD)accounted for 30.21%,34.73%,20.30%,8.81%and 5.95%of the area of the Puding County,respectively.The area of potential rocky desertification was the largest about 375 km2,and the area of severe rocky desertification was the smallest about 64 km2.(2)The time required for the operation of the neural network and the number of iterations can be sorted in the order from less to more:DE-BP<WOA-BP<GWO-BP<PSO-BP<BP.The initial BP neural network has the worst operation efficiency.The time for classification and verification is 98 seconds and the number of iterations is 39510.The operation efficiency of DE-BP is the highest,the time for classification and verification is only 62 seconds,and the number of iterations is 27682.(3)Based on the confusion matrix,an accuracy evaluation system including user accuracy,production accuracy,overall accuracy and evaluation indexes in kappa coefficient is constructed.In terms of user accuracy,GWO-BP has the highest accuracy in classifying NRD,MORD and SRD,of 86.39%,78.47%and 87.66%,respectively,and PSO-BP has the highest accuracy in classifying PRD and MRD,of 83.74%and 72.84%,respectively.In terms of production accuracy,GWO-BP has the highest accuracy in the classification of MRD and SRD,of 80.33%and 92.33%,respectively,and PSO-BP has the highest accuracy in the classification of NRD and PRD,of 92.00%and 68.67%,respectively,and WOA-BP has the highest accuracy in the classification of MORD,of 77.67%.In terms of overall classification accuracy,the initial BP neural network has the lowest accuracy of 74.73%,and GWO-BP has the highest accuracy of 81.40%.In terms of kappa coefficient,the initial BP neural network is the lowest,of 68.41%,and GWO-BP’s kappa coefficient is the highest of 76.75%.After synthesizing the four evaluation indexes,it can be seen that the classification results of GWO-BP neural network maintain higher consistency with the actual situation.