Design Of High-resolution Image Specific Shape Extraction System Based On BP Neural Network

With the construction of high-resolution image for Earth observation system,extracting all kinds of information from high-resolution satellite remote sensing image has become a research hotspot in remote sensing application.Different from the conventional classification and recognition of ground object in high-resolution image,extracting specific shapes in high-resolution image,only extracts ground object that has with the same edge shape in high-resolution image,rather than recognizes the types of objects.Namely,the culture with same edge shape is extracted as one category,and others are classified as another category.Specific shape means the shapes which have been specified in advance and that need to be extracted depending on practical requirements,including “round”,”rectangle”,and ”linear”.Extracting different shape features is important and meaningful for extracting and identifying artificial target ground object with contain shape.The topic starts with the application of the shape feature of the ground object,using BP neural network to extract and identify shapes of ground object,which has a certain guiding significance for high-resolution image retrieval research based on shapes' characteristics.This dissertation mainly includes two parts: algorithm research and system implementation.Algorithm research focus on extraction of characteristics of shapes and BP neural network algorithm.For extraction of shape features,firstly,pre-processing the high-resolution image with extraction positive shot correction,registration,fusion and cropping.And then using smooth filter for denoising.Next,Otsu two value,and edge Extraction using Canny operators and mathematical morphological operators,then extracting outer contour of the edge in Method of contour tracking.Finally,calculating shapes of high-resolution image which processed by Hu invariant moment and shape description Method of centrifugal,including "round","rectangle",and "linear".For BP neural network classification and extraction algorithm,constructing BP neural network structure satisfying classification requirement,getting Neural network parameters through training,and verifying the correctness of neural network classification after the actual test.For system implementation,designing specific shape extraction system of high-resolution image by invoking class libraries,interfaces,controls and programming under the Visual studio2010 development platform.BP neural network algorithm can classify and extract the irregular shape in High-resolution image by its strong generalization ability and nonlinear mapping capability,which can extract and identify general shape with Irregular shape of high-resolution image.It provides reference suggestions for research on shape extraction and classification technology of high-resolution image and software design.