Optical Images Assisted Ship Target Recognition Based On ISAR Images

The ISAR images have been widely used in ship target recognition,however,the scattering characteristics of ship targets are affected by many factors,thus the recognition accuracy is relatively low.Therefore,improving the recognition accuracy of ship ISAR images is a hot research topic.Optical images are more advantageous for human/unat power systems because of their high resolution,rich image information,high recognition rate,and more favorable features for human intervention.In actual situations,we usually can not obtain both the ISAR images and the optics images of ship targets.The working time of optical images is limited because of the weather and climate conditions(fog,smog,rain,clouds,etc.);even when the weather conditions are permitted,the optical imagse are still limited in imaging distance and affected by the light conditions.It is difficult to obtain an image of the target in the case where the light is insufficient.The ISAR is not only actively detected,but also can do long distance detection,and its resolution characteristics will not affected by the distance.The intensity of the light has no effect on the imaging results,and it can work in various climates and at various times.So in most cases,the ISAR images of the ship's target are the only images we can get.In this paper,we focuse on how to use the ISAR images to distinguish different targerts.We expect to use the existing information(ISAR images)to generate missing information(optical images),so as to improve the recognition accuracy of ship targets.The main contents include:Firstly,in this paper,we describe how the inverse aperture synthetic radar is imaged.The imaging mechanism of Range-Doppler is analyzed in detail,and then three different ship models are simulated.We analyze the three rotation forms of the ship targets,and describe the ralations between the forms and ISAR images.For the subsequent research,it is more convenient to preprocess the image and construct an image library of the ship ISAR image.Secondly,for the problem of image target recognition,this paper uses the network of convolutional neural networks to complete.This network is described in this paper.The purpose of each layer of convolutional neural network and its error-sensitive items are expounded in detail.This paper uses the convolutional neural network to classify the ship's ISAR image based on the image library.Identification,and the recognition accuracy was obtained,and the results were reasonably explained and analyzed.Finally,with regard to the use of existing information to generate missing information,this paper uses an anti-generation network to achieve this goal.Specifically,two types of networks are used.The first type is a Pix2 pix network.The ship's ISAR image is placed at the generator input of the network,and the corresponding optical image is placed at the input of the discriminator.The network is trained until the Nash balance is reached.At this time,the generator outputs the generated image close to the optical image,and then the generated image is combined with the ISAR image to obtain a new image.The convolutional neural network is also used to extract and classify the new image,and finally obtain the ship.The recognition rate of the target;the second type is the partial pattern confrontation generation network,whose generator input is the feature of the ISAR image,the authentication network contains the discriminator and the predictor,and the input of the discriminator in the authentication network is the optical image feature.The output is generated.The input of the predictor is a new feature after the fusion of the ISAR image feature and the generated feature.The predictor classifies the new feature and outputs the recognition accuracy of the ship.The simulation results show that the use of these two types of networks can improve the recognition effect of only the ship ISAR image.