Research On Surface Ship Targets Detection And Recognition Technology Based On Satellite Remote Sensing Images

Spatial remote sensing technology is one of the most important technologies to collect land information at present.Because of its unique space exploration advantages,it has a good development prospect at both military and civilian applications.The ocean is the main developing field in the world.The level of a country's marine technology is the sign to measure the level of development of the country.Due to the progress of remote sensing satellite technonogy,image recognition processing usually applied to extract informations,recognize and classify ship targets.It has great significance for fishery production,maritime treatment,marine security,strategic reconnaissance and other activities,in order to improve our ability to control the oceans.This paper will focus on the research of remote sensing image processing and recognition of ship targets.Firstly,the main categories of remote sensing image and the characteristics of sea surface remote sensing images are introduced in this paper.And it introduces the structure and operation principle of the integrated remote sensing system,and the operating process of the whole system.We list on some of the current remote sensing recognition methods which are introduced to explain the main process of the current target recognition method and analyze the shortcomings of these methods.Then we propose a new mathod based on deep learning theory to solve them.Through the analysis of the ship target on the remote sensing image,we filter images according to the image characteristics and enhance the contrast.We also need to eliminate clouds cover on the image processing to enhance the quality of the ship targets.Finally,we introduce some commonly used image segmentation methods and compared them to analysis their advantages.In order to separate ship targets from sensing images,we use graph portioning active contours for image segmentation on the ship target segmentation.Then,we should set up the data set according to the pre-processing remote sensing target image to build the deep learning network to identify the target.We first introduce several commonly used training data sets,explaining their structure and training picture characteristics,in order to learn how to make our own training set.Based on the structure of other training sets,we build our own surface ship target training set.We screen the training targets,and unify format processing package.At last finishing the training set and test set of the ship targets to convenient for network adjusting and testing.Finally,we introduce the principle and algorithm of deep learning theory.According tothe advantages of the algorithm and the main direction of the application,selected the convolution neural network to construct the target recognition network.According to the parameter characteristics of data set to build our convolutional neural network.In order to make convolutional neural network to achieve the requirements of anti distortion,detail identification sensitive and rotation invariance for ship targets,we puts forward the spatial transformation model convenienting for the network to meet the needs of remote sensing images.We train the convolution neural network which we build to identify the target and statistics recognition rate.The results of experiment prove that the network is feasible,and have good research and application value.