Design And Implementation Of The System For The Classification Of Military Targets Based On Google Map

In recent years, along with the continuous enhancement of informationization degree and the gradual complexity of electromagnetic environment in modern military struggle, recognizing and positioning detective targets accurately becomes more and more difficult. Therefore, how to recognize them quickly and accurately becomes the key factor. A great deal of information can be obtained through satellite pictures, but facing such enormous information, recognizing military target manually could not meet the needs of modern warfare. In order to achieve the goal of military target recognition, people devote to the research of making use of computer vision technology.On the basis of the character analysis of military target image from the Google Map, this article designed the corresponding classification algorithm, and realized the system of military target recognition with C#language programming based on the Visual Studio2005software development platform. The main completed work is as follows:(1) Realization of image interception. All satellite pictures in the article sample database are from GOOGLE MAP, the author intercepted10images for different categories of military targets, segmented the images, extracted images with the same size to analyse.(2) Grayscale processing of color images. Because of the satellite pictures are color images with actual environmental information, the author design the gray process to reduce the computation amounts in the process of images.(3) Main features selection and extraction of images. Based on the predecessors’ research and comparison on military target recognition algorithm, this article has realized chief features extraction of training sample with two parsing algorithm--principal constituent analysis and two-dimentional principal component analysis. It also obtained feature projecting space, and mapped the samples into the feature space, extracted useful feature vector, reduced dimensions of the initial data to made great reduce of subsequent operation calculation without creating affects of the classification results.(4) Realization of classifier training and pattern classification for military targets. In this paper, the training samples of military targets were entered K nearest neighbour (KNN) classifier, support vector machine (SVM) classifier and probabilistic neural network (PNN) classifier to do the classification test. The simulate experiment results of classifiers was statistical analysised through the leave-one method. Based on the statistical analyses of accuracies of the three classifiers, the SVM classifier has been chosen for its higher accuracy and stability. The system was tested by two and five categories of military targets, the accuracies were100%and75%respectively.(5) Initial realization of the model of military target classification recognition system. Connecting every fuction of modules including image segmentation, preprocessing, character choice and absortion, character classification in series, the system can realize integrally military target recognition fuction and obtain the ideal experimental result. Meanwhile, the author has realized system software sealing which enhanced the system’s portability, and enabled the developed system being installed and used on different computers, that is simple and convenient.