Abstract

Sugar maple logs (Acer saccharum Marsh.) were scanned using X-ray medical scanner in order to develop a classification procedure for this type of imagery (CT images). The classification procedure was required in order to separate sapwood from colored heartwood, knots, rot and bark. Five logs coming from a freshly cut tree (group 1) and three logs sampled from a sawmill yard (group 2) were chosen for this purpose. Two parametric supervised classification algorithms, a minimum distance (MDC) and maximum likelihood (MLC) ones, were qualitatively and quantitatively tested and the resulting thematic maps were filtered by a 5 × 5 median filter. The classification accuracy was evaluated with confusion matrix and Kappa analyses. Sapwood is known to be the key factor determining sugar maple lumber value. The sapwood identification accuracy was found to be 98.6% (MDC) and 97.2% (MLC) for group 1 and 80.7% (MDC) and 81.8% (MLC) for group 2, respectively. Misclassification of defects occurred mainly between knots and colored heartwood. The overall accuracy of classification was about 83.1% (MDC) and 82.6% (MLC) for group 1 and 76.4 (MDC) and 78.0% (MLC) for group 2, respectively. The Kappa value from MDC and MLC was 0.622 and 0.624 for group 1 and 0.440 and 0.470 for group 2, respectively. These Kappa values indicate the existence of strong and moderate degree of conformity between the reference data and the classification procedure for groups 1 and 2 of logs, respectively. Both classifiers show no statistically significant differences in their capability of separation of sapwood from the other classes. Nevertheless, as MLC accuracy for colored heartwood is higher than MDC accuracy in logs without bark (normal situation in sawmills), MLC appears at this stage as the better alternative for analysing CT images of sugar maple logs.