Research On Cloud Recognition And Segmentation Technology Of Remote Sensing Images Based On Deep Learning

With the rapid development of satellite sensors and data storage technologies,space observation technology has been greatly improved in recent years.At present,We can quickly acquire and store a large number of high-resolution remote sensing satellite images now and these remote sensing images have become an important means for us to observe and study the surface ecological environment and social environment.Due to the influence of weather factors,clouds always appear in remote sensing images,which reduces the quality of images and greatly affects the application and development of satellite remote sensing images.Faced with such a large number of satellite remote sensing images,it is important to recognize and segment the cloud quickly,efficiently and accurately to improve the image utilization.As an important branch of artificial intelligence,deep learning has achieved great success in the field of computer vision.In view of the above problems,we focuses on the cloud recognition and cloud segmentation techniques in remote sensing images with the help of deep learning methods in this paper.The main work of our paper is as follows:In the second chapter,we first expound the basic theory of deep learning neural network,and mainly introduce the related content of convolutional neural network,such as network structure,training process and common models.And then we analyze the relevant characteristics of cloud.The characteristics of the cloud layer and the underlying surface are extracted and compared from the aspects of radiation characteristics and texture characteristics,which lays a foundation for subsequent research.We study the problem of cloud recognition in remote sensing images in the third chapter.The face of massive image data,traditional methods rely on manual identification of the need to consume enormous energy and time.So we proposes a cloud recognition method based on convolutional neural network in this paper.Using the technical advantages of convolutional neural networks in feature extraction,the network model is trained by dataset images,and the deep features of the image are extracted to achieve high-precision recognition of the cloud image.Finally,we study the cloud segmentation problem of remote sensing images in the fourth chapter.We propose a cloud segmentation method based on full convolutional neural network.Firstly we introduce the full convolutional neural network and the residual neural network,and then we label the dataset images with Labelme.We build a full convolutional neural network model based on ResNet-50.Then we train the network model with training set data and test it on test set data.Finally,the effectiveness of the method on the cloud segmentation problem is verified.In general,deep learning has achieved satisfactory results in the cloud recognition and segmentation processing tasks in this paper,and has great application value in realizing the intelligent processing direction of remote sensing images.