Target classification via support vector machines

The area of automatic target classification has been a difficult problem fo r many years. Many approaches involve extracting information from the image ry through a variety of statistical filtering and sampling techniques, resu lting in a reduced dimension feature vector that is the input for a learnin g algorithm. We introduce the support vector machine (SVM) algorithm, which is a wide margin classifier that can provide reasonable results for sparse data sets and whose training speed can be nearly independent of feature ve ctor size. Therefore, we can avoid the feature extraction step and process the images directly. The SVM algorithm has the additional features that the re am few parameters to adjust and the solutions are unique for a given tra ining set. We apply SVM to a vehicle classification problem and compare the results to standard neural network approaches. We find that the SVM algori thm gives equivalent or higher correct classification results compared to n eural networks. (C) 2000 Society of Photo-Optical Instrumentation Engineers . [S0091-3286(00)02703-3].