A new method for defect detection in lumber images: optimising the energy model by an irregular parametric genetic approach

Abstract

Accurate detection of surface defects plays a vital role in automated analysis of lumber quality in the wood industries. A new method, based on a genetic optimisation of energy model, is introduced here for defect detection in lumber images. In this method, a hypothesis testing framework is defined, first to separate defects from natural tissue of lumber. Then, the boundary of lumber is estimated by a decision function based on the energy optimisation method which is driven by an irregular parametric genetic approach. Performance of the proposed algorithm was evaluated on real captured lumber images containing several types of surface defects. The results demonstrate that the proposed method extracts the defects approximately 4.4% better than its alternatives while at the same time decreasing false detections by approximately 5.4%. The results obtained also show considerable improvements in accuracy and precision of the proposed method compared to other examined methods, especially when a low false detection (i.e. at least 10%) is desired.

Defect Detection of Industry Wood Veneer Based on NAS and Multi-Channel Mask R-CNN

“…By comparing the results of all models, we found that the number of filters strongly affects the accuracy of the model. Fewer filters can effectively decrease the FLOPs but the accuracy also decreases, which is not acceptable for wood defect detection in our study, even though the accuracy is more than 90% [47]. The kernel size also slightly affects the accuracy of the model but it has a strong effect on the FLOPs, as found by comparing Models 4 and 5.…”