Search

Automatic change detection methods for identifying the changes of serial MR images taken at di erent timesare of great interest to radiologists. The majority of existing change detection methods in medical imaging,and those of brain images in particular, include many preprocessing steps and rely mostly on statistical analysisof MRI scans. Although most methods utilize registration software, tissue classi cation remains a di cultand overwhelming task. Recently, dictionary learning techniques are used in many areas of image processing,such as image surveillance, face recognition, remote sensing, and medical imaging. In this paper we presentthe Eigen-Block Change Detection algorithm (EigenBlockCD). It performs local registration and identi es thechanges between consecutive MR images of the brain. Blocks of pixels from baseline scan are used to train localdictionaries that are then used to detect changes in the follow-up scan. We use PCA to reduce the dimensionalityof the local dictionaries and the redundancy of data. Choosing the appropriate distance measure signi cantlya ects the performance of our algorithm. We examine the di erences between L

We used a change detection approach based on support vector machine (SVM) to analyze two remotely sensed images in order to analyze urban landscape change on high-dense urban use (HDU), medium-dense urban use (MDU) and low-dense urban use (LDU) in Kunming, China. These two images were subset of a TM image acquired on 16 August 1992 and an ETM+ image acquired on 2 November 2000, respectively. First, we used SVM to classify each subset into HDU, MDU, and LDU. Then, we compared the label values of classified data pixel by pixel to analyze urban landscape changes. In order to obtain high quality training data under the circumstance that existing classification products of sampling area were not available, we proposed a second sampling method to assure obtaining satisfactory training data. The kernel function of SVM was radial basis function (RBF). Optimal model with the best penalty parameter C and the kernel parameter gamma was obtained through training samples. We tested the approach in three sites: northern Kunming, southern Kunming and entire Kunming. Results indicate that the overall urban use has substantially increased during 1992- 2000, while the substantial growth in high-density urban use was achieved at the cost of low-density urban use and partially medium- density urban use.

Entities in the real world have non-spatial attributes, as well as spatial and temporal features. A spatial-temporal data model aims at describing appropriately these intrinsic characteristics within the entities and model them on a conceptual level so that the model can present both static information and dynamic information that occurs over time. In this paper, we devise a novel spatial-temporal data model which is based on Geodatabase. The model employs object-oriented analysis method, combining object concept with event. The entity is defined as a feature class encapsulating attributes and operations. The operations detect change and store the changes automatically in a historic database in Geodatabase. Furthermore, the model takes advantage of the existing strengths of the relational database at the bottom level of Geodatabase, such as trigger and constraint, to monitor events on the attributes or locations and respond to the events correctly. A case of geographic database for Kunming municipal sewerage geographic information system is implemented by the model. The database reveals excellent performance on managing data and tracking the details of change. It provides a perfect data platform for querying, recurring history and predicting the trend of future. The instance demonstrates the spatial-temporal data model is efficient and practicable.

Sewerage network is an important part of municipal infrastructure for a city. Obstruction of sewer causes street flooding and affects people's daily life directly. To investigate reasons why some sewage pipes are blocked frequently in Kunming, China, we employ spatial analysis and data mining technology to analyze the data on the basis of a municipal sewerage geographic information system of the city. In the GIS, all of map layers and attribute tables are organized and saved in a relational database with Geodatabase model. First, we combined SQL attribute query with spatial location query to find out the sewage pipes that are blocked frequently. Then, we carried out buffer analysis and intersect analysis on the layers of the frequently-blocked pipes and buildings along the streets to extract buildings that are close to these frequently-blocked pipes. Joining the buildings in the buffer scope and the frequently-blocked pipes forms a big table prepared for spatial data mining. We used Apriori algorithm to mine spatial association rules from the data in the big table in order to search implicit reasons of obstruction of the pipes. The results from data mining indicate that strong spatial and non-spatial associate rules exist between the obstruction and restaurants in the buildings, as well as attribute slopes and diameters of these sewage pipes.

Lagragian support vector machine (LSVM) is a linearly convergent Lagrangian, which is obtained by reformulating the quadratic program of a standard linear support vector machine. To investigate the performance of the classifier working on multispectral images with LSVM as optimizer, we devise a new test based on LSVMs for classifying multispectral data in this work. First of all, data are preprocessed. To acquire the optimum bands for image classification, multispectral image is mapped into a two-dimensional feature space to inspect the bands with redundant spectral information. These extracted data acquired through the feature selection is named data group B relative to the original data group A for a purpose of comparison. Then, to classify multiclass problem, binary classification is extended to multiclass classification by pairwise method. Secondly, two groups of data are trained to find models. In this phase, optimal C values are chosen carefully through trials with different values. Then, classifiers based on LSVMs with optimal C values are used to yield optimal separating hyperplane (OSH). Lastly, in prediction phase, the two groups of data are inputted respectively into each classifier for testing. These classifiers include ones with linear kernel and ones with polynomial kernel of degree 2. The results of the experiment reveal that classifiers with LSVMs as an optimizer have excellent performances with both linear kernel and polynomial kernel of degree 2. Bias caused by the differentia of the two groups of data is not obvious.