AN INFORMATION EXTRACTION METHOD OF SUBURBAN INDUSTRIAL LAND USING IMPROVED DENSENET NETWORK IN REMOTE SENSING IMAGES

摘要

Aiming at the problem that traditional classifi-cation algorithm and shallow learning algorithm are ban industrial land in remote sensing images.In thispaper,a method of extracting industrial land infor- proved DenseNet network is proposed.Firstly,thedense neural network (DenseNet) is improved,and the SE block is embedded into DenseNet.Using thecharacteristics of DenseNet feature reuse and effi-cient information flow,the effect of SE block to ex-tract effective remote sensing scene image featuresis improved.Then,the learning rate is set in the wayof degenerate learning rate.At the beginning oftraining,the learning rate is used to accelerate train-ing of network model,and learning rate is reduced toseek optimal solution and improve accuracy of infor-mation extraction.Finally,in order to prevent thelack of fitting caused by lack of data,translation,ro-tation and other operations are used to expand train-ing data set.Experiments are carried out in Tensor-flowframework to demonstrate the proposedmethod.The results show that the proposed methodcan achieve fast convergence,high extraction effi-ciency,and the accuracy and loss are better thanother comparison methods.