Research On Information Extraction Of Forest Swamp And Shrub Swamp Based On Fully Convolutional Neural Network

The successful launch of high-resolution remote sensing satellites represents a new era in the acquisition and processing of geospatial data.It not only improves the accuracy of earth observation,but also accelerates the update speed of geospatial data.Expanded the scope and depth of remote sensing applications.Compared with low-and mediumresolution remote sensing images,high-resolution remote sensing images have the characteristics of high temporal resolution,high spatial resolution,multiple imaging spectral bands,and short revisit periods,which can describe features of ground objects in more detail.Information,such as detailed information,complexity of features.Therefore,high-resolution remote sensing images are more and more widely used in scenes such as image classification,dynamic changes of ground features,and environmental monitoring.For the field of image recognition,convolutional neural network,as one of the representative algorithms of deep learning,has received more and more attention,and has made domestic and foreign scholars have made breakthroughs in the field of image recognition.The central idea is to combine the model’s local receptive field,weight sharing,pooling operation,etc.,to reasonably and effectively reduce the number of parameters in the network model,alleviate the problem of overfitting,and achieve the goal of optimizing the network model.The fully convolutional neural network model is compared with the convolutional neural network.The latter is the basis of the former.The former transforms the fully connected layer in the latter model into a convolutional layer.Based on the full convolutional neural network model,this paper extracts the wetland type information of forest swamps and shrub swamps from the high-resolution remote sensing images of the Dazhanhe National Nature Reserve,and uses the full convolutional neural network model to better identify images Features,excavate feature information.The main research contents of this paper include:(1)Carry out remote sensing image classification research on maximum likelihood classification method,iterative self-organizing classification method,object-oriented classification method,and full convolutional neural network model.The convolutional neural network analyzes,verifies the accuracy of the classification results of the remote sensing images of the above four methods,and finally compares and analyzes the obtained results.(2)The data used include the 2018 Gaofen-2(GF-2)multispectral remote sensing image data of the study area and the panchromatic data corresponding to the 2 scenes.In order to determine the best segmentation effect of the image,the forest swamp and The information of the two types of wetlands in shrubs and swamps is segmented by objectoriented multi-scale,and the images are segmented with different segmentation scales,and the segmentation results are compared.(3)Input the segmentation results of the two types of wetland information of forest swamp and shrub swamp with the determined segmentation scale into the full convolutional neural network model to realize automatic image recognition and feature information extraction.The accuracy of the classification results of the model is verified,and the results show that the accuracy of the classification results of the model is relatively high,indicating that the model used in this article has a good generalization ability for the learning of data features.The research results show that compared with the traditional classification method,the full convolutional neural network,as an’end-to-end’ network structure model,can dig out the feature information of features in more detail,and has the highest accuracy in image recognition and classification.The remote sensing image classification method of the fully convolutional neural network model can obtain higher classification accuracy.