WD Detector: deep learning-based hybrid sensor design for wood defect detection

Abstract

The fast-growing human demands in the world are leading to the expansion of industrialization. As wooden materials are increasingly used in industrial settings, detecting defects in wood has become crucial. Wood defects adversely affect the quality and durability of materials. A wood defect detection method, named WD Detector, is proposed in this study to identify wood defects. There are 18,284 defective wood surface images and 1,992 undefect wood images in a dataset of 20,276 wood images used for wood defect detection. 12 different classical machine learning algorithms are used to classify wood defects after extracting features from images with various CNNs and transfer learning approaches. In this study, feature extraction is performed by training the Xception CNN model. Once the features are extracted, classical machine learning algorithms are used to classify the wood defects. For the first time, a deep learning-based hybrid sensor design has been implemented on this dataset for wood defect detection. WD Detector achieved 99.32% accuracy in detecting wood surface defects using the new method. The success of this study’s method in detecting wood defects is believed to pave the way for future studies.