IMARS is a distributed Hadoop implementation of a Robust Subspace Bagging ensemble Support Vector Machine (SVM) prediction model for classification of imagery data. IBM’s Multimedia Analysis and Retrieval System (IMARS) is used to train the data. These data are manually categorized for various land-use types to ensure that they are correctly identified in training data. These tiles are units of parallelization for Hadoop implementation. Training data are obtained from GeoEye public domain, and the imagery is divided into 128 × 128 pixel size tiles with 0.5 m resolution. To determine land use, semantic taxonomy categories such as vegetation, building, pavements, etc. (2011) present a Hadoop-based distributed computing architecture for large-scale land-use identification from satellite imagery. High-resolution imagery is also used during to natural disasters such as floods, volcanoes, and severe droughts to look at impacts and damage. Land-use data are used extensively for urban planning. It can be used to identify different areas by the type of land use. Complications involved in the representation of training/output data.Not the most efficient method to find some optima, rather than global.Computation or development of the scoring function is nontrivial.Good at refining irrelevant and noisy features selected for classification.Efficient search method for a complex problem space.Can handle large, complex, nondifferentiable and multimodal spaces.Always finds a “good” solution (not always the best solution).Can be used in feature classification and feature selection.Precise solutions are not obtained if the direction of decision is not clear.Prior knowledge is very important to get good results.Different stochastic relationships can be identified to describe properties. ![]() Difficult to determine optimal parameters when training data is not linearly separable.Training is slow compared to Bayes and decision trees.Easy to control the complexity of decision rule and frequency of error.Computational complexity reduced to a quadratic optimization problem. ![]()
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |