Research On Deep Convolution Neural Network In Wetland Type Information Extraction

High-resolution remote sensing images can describe the details and complexity of ground objects in detail.Compared with low-and medium-resolution remote sensing images,high-resolution remote sensing images have better performance in texture,shape,spectral characteristics and so on.Therefore,they are widely used in the fields of ground objects classification,surface observation,dynamic monitoring of natural resources and so on.However,the pixel-based classification method and shallow machine learning algorithm for medium and low-resolution remote sensing images can not meet the classification requirements of high-resolution remote sensing images.How to improve the efficiency and accuracy of high-resolution remote sensing image classification has become a research hotspot.With the continuous improvement of computer performance and the in-depth development of artificial intelligence,a large number of scholars have applied deep learning methods to various fields.The research shows that the deep learning method can effectively solve the frontier problems such as classification and discrimination of massive images,which provides a reliable support for the classification of high resolution remote sensing images based on the deep learning method.In recent years,deep convolution neural network as a deep learning model has achieved a major breakthrough in the field of image recognition.Its core idea is to use the combination of local receptive field,weight sharing,and pooling operations to optimize the network so that it has a certain degree of invariance such as translation,scaling,and distortion.Based on the depth convolution neural network model,this paper extracts the wetland type information from high resolution remote sensing images,better expresses the depth characteristics,and excavates the terrain information.The main research contents of this paper include:(1)Research on traditional supervised and unsupervised classification methods for remote sensing images;the deep convolution neural network is analyzed from the aspects of network structure,parameter setting,etc;the traditional classification method iscompared with the deep convolution neural network.(2)Taking Heilongjiang Gongbiela River National Nature Reserve as the research area,the multi-spectral remote sensing image data of GF-2 was used in June 2018.The information of wetland type is divided into six kinds:20,30,40,50,100 and 150,and the best segmentation effect is determined.(3)By using the deep convolution neural network model to construct the deep network structure and adjust the model parameters,the wetland type information can be automatically identified and extracted.Through automatic classification of test samples other than training set and verification set,higher accuracy classification results are obtained,which reflects the generalization ability of the model for data feature learning.The results show that,as a deep structure model,the deep convolution neural network can mine the information of ground objects more deeply and express the depth characteristics.By combining multi-scale segmentation of remote sensing image with depth convolution neural network,image information can be automatically recognized and extracted,which can achieve higher classification accuracy and efficiency.The feasibility and reliability of applying convolution neural network to high resolution remote sensing image classification are proved.