PATIERN RECOGNITION OF LENTIL SEEDS USING MACHINE VISION AND NEURAL NETWORKS
| dc.contributor.advisor | Wood, H. | |
| dc.contributor.advisor | Sokhansanj, S. | |
| dc.creator | Winter, Philip W. | |
| dc.date.accessioned | 2017-08-25T17:36:29Z | |
| dc.date.available | 2017-08-25T17:36:29Z | |
| dc.date.issued | 1997 | |
| dc.date.submitted | Spring 1997 | en_US |
| dc.description.abstract | Machine vision systems, computer-based visual pattern recognition, has been proven a useful tool in many industries, mostly in the . automated inspection of manufactured items. In the production of grain in the agricultural industry, a similar inspection process occurs for the quality inspection of the seeds. This current grading process requires highly trained human inspectors and entails many repetitive measurements which sometimes leads to inconsistencies. A machine vision tool would be very beneficial in the grain inspection process in improving efficiency and objectivity. Lentil seeds are presently graded visually based on the Canadian Grain Commission's established criteria: seed colour, size, damage, and foreign material. In this work, a machine vision system was developed to differentiate commercial samples of lentils (Laird variety) into three classes of "good", "discoloured", and "broken and peeled". A neural network was used to analyse 21 morphological and colour features of a sample and provide an output between -1 and 1. Two architectures of neural networks were tested: the multi-layer neural network (MLNN) and the multi-structure neural network (MSNN). The best system performance obtained was as follows: 94% success in recognition of good lentils, 97% for discoloured lentils, and 95% for broken and peeled lentils, for an overall recognition rate of 95.6%. This result was obtained using an MLNN network. The MSNN network gave comparable results with an overall accuracy rate of 95%. The features used were also analysed for their contribution to the pattern recognition problem. They were ranked using the accumulated correlation coefficient (ACC) method and by eigensystem analysis. A reduced set of optimal features was used to train an MLNN, with an overall accuracy of 93%. Finally, the projected areas of the lentils, and the mass of the lentils were correlated and a linear equation was developed so that results of the pattern recognition could be presented in the form of a percentage by weight, as is the industry standard. | en_US |
| dc.identifier.uri | http://hdl.handle.net/10388/8062 | |
| dc.title | PATIERN RECOGNITION OF LENTIL SEEDS USING MACHINE VISION AND NEURAL NETWORKS | en_US |
| dc.type.genre | Thesis | en_US |
| thesis.degree.department | Electrical and Computer Engineering | en_US |
| thesis.degree.discipline | Electrical Engineering | en_US |
| thesis.degree.grantor | University of Saskatchewan | en_US |
| thesis.degree.level | Masters | en_US |
| thesis.degree.name | Master of Science (M.Sc.) | en_US |