Algorithm Research For Aircraft Recognition And Classification In Remote Sensing Images

With rapidly changing of modern environment, real-time and near real-time information and the immediate judgment of the nature and value of information are necessary. The intelligence which is delayed too long can not effectively help combat activities, but may delay fighters. Air Force has the world's most advanced intelligence gathering, processing equipment and the means of real-time space-ground transmission, But they find that the crucial flaws of aerial reconnaissance intelligence in the tactical support system is the speed which Image Intelligence offered can not meet the requirements of modern warfare, when they reviewed the lessons of Gulf War aerial reconnaissance intelligence.To meet need of the future high-tech wars, military intelligence software has been extensively applied. Aerial imagery Automatic Target Recognition is an important research topics in image processing and pattern recognition area and has high academic and practical value. As the aircraft is an important military target and one of the main targets of reconnaissance and attack in the military field, it's important for accomplishing target strike mission to judge the type, condition and quantity of ground parked aircraft by using aviation images and identify weapons which the aircraft carried.In the research field of the identification and location of military targets, the military intelligence analysis and objective assessment methods of U.S. force based on Remote sensing images has integrated with the overall lunch of cruise missiles and the remote attack security system of combat aircraft and becomes further consummate with the advance of Image acquisition means,analysis and processing technology and the development of computer technology. The application has proved such a system has played an important role. Former United States Department of Defense Mapping Agency (DMA) using high-resolution infrared images of KH-11 spy satellite, NASA SAR, SPOT, TM images, as well as military aircraft reconnaissance images to real-time monitoring and analysis the military objectives and the development of the situation and transmit data to missiles and aircraft which execute continuous attack mission. Technology level of such systems of U.S. force is at a new height when NATO in 1999 against the Federal Republic of Yugoslavia 78days of massive air strikes. This increased performance focus in objective analysis of data sources. The most notable feature is using diversified aerospace and aviation image data synthetically and consequently constitute all-weather and all-round surveillance to the target. Space data are mainly offered by "Lacrosse" radar satellites and improved KH spy series based on satellite. Aviation data are mainly offered by Special scout such as TR-1,U-2,RC-135 etc. Allies of U.S. in the West also provides a variety of data. For example, the images offered by France "Apollo" (Helios) 1st reconnaissance satellites and "Mirage" IVP reconnaissance scout play a positive role in for the assessment of the effectiveness of the air strikes.Functions of the computer system for objective assessment in particular the software system has become more powerful. Firstly, a higher intelligent image recognition system mainly by automatic identification subsidiarily by human-computer interaction identification has been put into use. This not only enables the computer to process and analyze image data capacity of the pace has quickened, but also makes the system can be more oriented to the battlefield and regional clusters targets integrated assessment mission. Secondly, in target assessment and drawing aspect, we can joint spatial data analysis functions of theater of operations military geography information system and make full use of corresponding mature business software system, thus realize the automatization of this part basically. Thirdly, with the development and application of virtual battlefield 3D scene software, object recognition effect becomes more lively,detailed and vivid. Typical is mainly directed against military objectives such as bridges, buildings, ports, airports, etc.At present, China's army in the military automatic target recognition area, using multi-spectral, multi-temporal satellite remote sensing images and the spectral properties of target to identify objectives. Typical is mainly directed against military objectives such as bridges, buildings, ports, airports, etc. This does not meet the requirements of the real-time war, accuracy and reliability. Especially in the combination of military objective identification and location, China has no means perfect application system. This article tries to base on the exploration in the aerial imagery automatic target recognition and target detection technology, meets the requirement of our military to win the future information war, makes a detailed study on the aircraft target Recognition of the visible light images in remotesensing images, and finally complete the automatic mode recognition system. The main contents of this paper are:(1) Image preprocessing of wavelet theory remove noise. In this paper, using a frame based on wavelet denoising image preprocessing, Based on the scale of the basic wavelet "window of the center" displacement over sampling and the wavelet tribe, It wavelet (with the exception of the scale outside the constant factor) as its subset. Displacement with the same choice of wavelet transform, thus achieving the signal denoising, filtering.(2) System improved regional segmentation and fuzzy poor video image segmentation approach to the target area.Focused on solving complex aerial images of the background object detection and false alarm excluded.(3) Target Feature Extraction of three invariant moment Hu moment, wavelet and fuzzy moment affine invariant moments earlier, wavelet moment invariants and fuzzy affine invariant moments aerial images of the aircraft target feature extraction and identification of excellent results fruit. We will invariant moments and fuzzy affine invariant moment for target identification, the establishment of a model invariant pattern recognition. Using a method to make such a pattern recognition model with location, size and rotation invariant; Experiments show that the model is not only good recognition accuracy but also the good performance of the defect and anti-jamming capability.(4) Categories in the design, we wavelet-based BP neural network classification methods of feature selection. First, good stability, the ability of anti-rough classification of features, then strike a corresponding category and the characteristics of the eigenvector into classifier identified. This approach greatly enhance the whole identification process more efficient.