Research On Military Bastion Target Features And Recognition Technology In Remote Sensing Images Based On Structural Model

Military targets arrays' features, especially the spatial distribution features in remote sensing images are studied in this paper.Firstly, the characteristics of resource data are discussed. And then, the fuzzy theory and the graph theory are confirmed to be the main research theories of this paper.Some color invariant features which invariant with the lighting condition, the view point position, the Fresnel reflectance coefficient of object's material are studied based on the Kubelka-Munk physical illumination model in chapter 2. The spectrum energy distribution structure is studied by using Gaussian color model, and a color invariance calculation method from RGB component of images is proposed. The color invariant features are used for fast segmentation of colored remote sensing, or multi-spectral images. A fast cluster number adaptive fuzzy C mean algorithm based on feature quantitative is proposed for extracting the sub-target blob efficiently.The military targets array is customized by ARG model in chapter3. By investigating the "path" concept of graph theory, the traditional ARG model is extended to ARmG model, and a multi-side relationship system Rmof the nodes based on the path description is established. As a result, ARmG model is used for analyzing the spatial distribution relationship among targets by studying the shape of the path that connects tagets. A multi-layered ARmG (mARmG) model is proposed for analyzing the multi-scale structure of targets' relationships. The model has the form as G(G,E}, in which the node can be considered as a sub-graph. And then, the relationshipattributes of mARmG model are defined by fuzzy set. The mARmG model is extended to mARmG model, and becomes the data resources for fuzzy information integration.The geometric attributes of target blobs are studied in chapter4. A distinct feature of targets array is presented based on binary distance relationship of nodes for recognizing the side by side arrays. And then, the Hu invariant moments are modified for decribing the spatial distribution shape of discrete point sets. Some useful features for telling the lining, the circling, the grouping distribution structures of point sets are presented. And then, a comparation study of fuzzytopology is carried out. A fast fuzzytoplogy caculation method which is suitable for processing large remote sensing images based on the fuzzytoplogy envirnment morphology transfer alogorithm is presented. The presented method is more efficient that Bloch's method. At last, some formulae for caculating "Near" and "Surround" fuzzytoplogy more intuitively are presented.A novel targets' spatial distribution structure detection algorithm which we call as max spatial regularity detection algorithm (MSR detection algorithm)based on fuzzy information integration is proposed in chapter5. The aim of algorithm is to get a spanning tree constraint by the paths that have max spatial relationship regularity. By utilizing the minimum cast spanning tree growth mechanism of Prim algorithm, MSR integrates fuzzy spatial information byestimating the fuzzy spatial relationship regularity among neighboring nodes by back searching the path in the found tree and forward detecting the nearest nodes, and detects the regular spatial relationship among targets initiatively as its can. The detected MSR spanning tree can be considered as the fuzzy evaluation of the regularity of array's spatial distribution structure, and its' some features can be used to detect targets' array. The experiments show that the algorithm iseffective and stable. The complexity of algorithm is maintaining O(N2), where N is thenumber of targets in the array.The problems and content needed further research are pointed out in chapter 6.