The k-nearest neighbor method for automatic identification of wood products
Résumé
The complete follow-up of products from its origin to its final use has been strongly developed in the last years. In wood industries, the implementation of ordinary identification systems presents implantation problems, mainly because of the extremely variable nature of the material and the particular features of the manufacturing process. In order to allow follow-up of products, new solutions have been considered by the use of non-destructive control techniques. We implemented an identification system, where each product is considered as unique with unique features. For this research we use a microwave sensor to obtain an intrinsic signal of a wood piece and then perform its identification. The objective is to determine a process for identifying signals. The algorithm for pattern identification developed here is based on the k-nearest neighbor method. To increase performance of this algorithm, the signals are pre-treated. Signals already recorded in the database are positioned in the space to n-dimensions. When a product is to be identified it will also see its signal positioned; the nearest signal having as a reference the distance criterion will be located. The identifying algorithm developed shows a percentage of error of 1,5%.