Research On The Method Of The Target Recognition And Orientation Based On The Neural Networks

The dissertation is the study of the target recognition and orientation which has change of angle,translation and scale under the simple background,based on the neural networks.The content includes in the fuzzy segment,BP neural networks and principal component analysis.The method of the whole image recognized is present through neural networks. In order to find the target the pattern method need search each line and each row, it is cost more than present the method. When the targe's scale is two-part of the image or the target is in the center of the image, the recogniton and the orientation is implemented together by neural networks. It is necessary to compute the target position when the target is smaller than half of background and is not center. So the dissertation not only discusses the neural networks recognition method but also studies the orientation method. This thesis mainly consists of the following parts.Multilyer feedforward networks and the BP neural network algorithm are analied. The selection of hide layer and number of nerve cell of hide layer are introduced. The weight value and paramer are optimized. The result is train cost is smaller and train effect is better.The input eigenvector is important for the net studied successfully. So the two eigenvectors are discussed. One is the moment invariant, the other is principal component.The excellence and disadvantage are discussed.The dissertation presents extracting the moment invariant of processed image and a kind of fuzzy segment way during processing image.The experimental results demonstrate the proposed approach is effective.The principal componet can be gotten from the gray image. It is difficulty to getting the principal componet. So the principal componet network is presented to get the principal componet.