Authors: Tushar Jain; Meenu; H.K. Sardana
Addresses: Mechanical Engineering Department, MIET, Meerut, India ' Mechanical Engineering Departments, NIT, Kurukshetra, India ' Central Scientific Instrument Organisation, Chandigarh, India
Abstract: Automated recognition of mechanical parts is a task in manufacturing that has been automated at a comparatively slow pace. Nearly all of the existing object recognition systems, with the exception of very few experimental systems have been designed to recognise a single object. In this paper, this problem is solved in a great manner so that the same process can handle different 2D recognition applications. Colour images are used during object recognition. The Fourier descriptor method has been adopted for recognition of mechanical parts. This method recognises an object by extraction of features from an object image. The objects may be classified using artificial neural network (ANN). For training and testing in either case, the features are extracted by presenting the object in different orientations. A feed forward neural network structure that learns the characteristics of the training data through the back-propagation learning algorithm is employed. The emphasis is put on the choice of network architecture and setting of different parameters. The study also considers the effects of various user-defined parameters and noting their effect on classification accuracy. The effect of orientation angle of the object and sample size on overall accuracy is also considered on the used classifier.
Keywords: automated recognition; mechanical parts; object recognition systems; image processing; artificial neural network; ANN; intelligent machines; robotics.
International Journal of Intelligent Machines and Robotics, 2018 Vol.1 No.2, pp.122 - 132
Received: 01 Jun 2017
Accepted: 08 Sep 2017
Published online: 26 Sep 2018 *