The Neural Computing Methods For Spatial Data Analysis

Spatial data also known as geospatial data or geographic information it is the data or information that identifies the geographic location of features and boundaries on Earth, such as natural or constructed features, oceans, cities, roads and more. Spatial data is usually stored as coordinates and topology, and is data that can be mapped. Spatial data is often accessed, manipulated or analyzed through Geographic Information Systems (GIS). The analysis of spatial data can be found in many fields, such as resource planning, animal population migration analysis, disease diffusion analysis, crime analysis, road network analysis, earthquake prediction and remote sensing image processing etc. With the development of science and technology, the spatial data volume has an explosive growth tend. The rapid growth of spatial data brought severe challenges to traditional methods of spatial data analysis.In order to process spatial data more efficiently, this paper mainly focus on the research of spatial data processing by using neural computational methods. This dissertation studies how to use Pulse Couple Neural Network (PCNN) to process spatial data.The main results are as follows:1. A spatial point-data reduction method by using PCNN is proposed. A modified PCNN model is designed to search data boundary for spatial datasets. The method is a general method, which can be used to improve several different spatial data analysis tasks. On the other hand, the method can filter noises effectively, and achieve reasonable generalization accuracy.2. A sparse LS-SVM algorithm is proposed. LS-SVM is a modified SVM, which only needs to solve a set of linear equations instead of a quadratic optimization problem. However, the sparseness is lost because LS-SVM makes use of anε-sensitive cost function. To impose sparseness to LS-SVM solution, this dissertation proposes a direct method to impose sparseness to LS-SVM, which is done by pre-selecting some more significant data points as candidate support vectors for LS-SVM. This method is time-saving and can effectively improve the sparseness of LS-SVM. 3. Two modified PCNN models are proposed to solve the shortest path problem in road networks. The first model, called CPCNN, has a special On-forward / Off-backward competitive mechanism. By using of the competitive behavior, the model can inhibit some useless firing events and encourage those important neurons. As a result, CPCNN can reduce search space significantly, and it can search shortest path more efficiently. Another model is called DSPCNN, which has two firing sources and two Autowaves propagates in parallel to search the shortest path. The DSPCNN is also faster than some existing PCNNs.4. A motion segmentation method using non-uniform sampling based density clustering is proposed. In this method, the frame difference image is sampled by the proposed non-uniform sampling scheme, and then transformed into a spatial dataset. Next, the segmentation result is obtained by using density clustering method for spatial data.This dissertation focuses on how to process spatial data more efficiently. The results show our researches can improve some spatial data analysis tasks, which can be used in some application fields.