Resumen de Identificación del cilindro nudoso en imágenes TC de trozas podadas de pinus radiata utilizando redes neuronales artificiales
Gerson Rojas Espinoza, Óscar Ortiz Iribarren
The feasibility of identifying Knotty core in images of X-ray computed tomography (CT)of pruned radiata pine logs (Pinus radiata D. Don), was evaluated using a supervised classificationmethod based on artificial neural networks (ANN). The classification process also considers theidentification of the clear wood and knots. Thirty pruned radiata pine logs were scanned in a multislicescanner medical X-ray, where the resulting CT images were obtained every 5 mm. A total of270 CT images were classified using the ANN, and the resulting thematic maps were filtered with amedian filter of 7 x 7. The accuracy of the classification process of the CT images was obtained froma confusion matrix and Kappa statistics. The results indicated that the Knotty core can be identifiedand separated with an accuracy of 92.7%, while for the overall accuracy was obtained a value of85.0%. After filtering thematic maps, the precision values increased to 96.3% and 92.3% for thedefective core and overall accuracy, respectively. Kappa values were 0.607 and 0.764 for thematicmaps and thematic maps filtered, respectively. These values indicate that there is a strong degree ofagreement between reference data and classification process. The results suggest that it is feasible toapply artificial neural networks as classification procedure to identify the Knotty core in CT imagesof pruned radiata pine logs.
