A supervised approach to automatic density and foreign-object detection in particleboard paving quality control

Abstract

ABSTRACT Density detection and foreign object detection are very significant to ensure the quality of particleboard production and are important links in the quality control of pavement. This study analyzes the density uniformity of the paving process by average surface density, horizontal and vertical zoned average surface density based on CT values detected by X-rays, and studies the parts and causes of density non-uniformity. A foreign object detection scheme of alarming and then detecting is proposed to detect the presence of foreign objects in paving boards by MobileNetV3. Classification and localization of defects are performed by the improved YOLOv8 model with a multi-window sliding detection method, which introduces the backbone network of MobileNetV3 to simplify the network structure, and enhances the feature extraction capability and the detection of small targets by improving the C2f of YOLOv8 and increasing the detection branches. The experimental results show that the detection precision of the foreign object alarm is 96.72%, the recall rate is 95.80%, and the mAP value is 95.65%. In the defect slice dataset, the improved YOLOv8 defect detection algorithm has a detection precision of 97.96%, a recall rate of 97.60%, and a mAP value of 89.55%.