Automatic Target Recognition System Using Self Learning Algorithms

Abstract

The present era of computational automation is significantly contributed to by machine vision, newlinewhich carries the computational intelligence-based approach to make the solution more newlineaccurate and reliable. The increased complexity of the automation environment in various newlineapplications demands knowledge-based processing rather than the conventional comparisonbased newlineapproach. The detection of targets is considered as one of the prime objectives in newlinemachine vision and can be seen in different areas of application like automation of medical newlineimage analysis, surveillance, autonomous vehicles, and industrial automation, to name a few. newlineMost applications imposed constraints on the level of recognition accuracy, variability in the newlinepresence of noise, and computational cost when detecting targets. The detection of a target newlineneeds to define the model of a target and can be fulfilled under two categories: (i) feature based newlinemodelling where the exclusively low-level target features are extracted and a model defined newlinefrom them for final detection; (ii) knowledge-based modelling where the mapping of features newlinetakes place from image information to other numeric ranges under the adaptive environment. newlineThe approach of feature-based modelling carried a lack of specificity and a larger newlinecomputational cost, while knowledge-based modelling needed intelligence. The proposed newlinework fundamentally acquires the knowledge mapping based approach by the development of newlinea self-learning environment using the neural network, evolutionary computation, and support newlinevector machine. The considered targets were selected over the application of advanced driving newlineassistance in the detection of traffic signs alongside the road and the detection of road lane newlinealong with the availability of road clearance by detection of objects in the near vision distance newlineon the road. The recognition of different traffic sign images was done in the three cascaded newlinemodels carrying the dimensionality reduction, neural network model of the learning newlineenvironment and the correlation based distanc