Research on Visualization Method of Edge Banding Appear-ance Quality Based on YOLOv7

Abstract

Edge banding plays a critical role in contemporary panel furniture production, as its quality significantly affects the aesthetic appeal of furniture. However, the current method of quality inspection and classification relies primarily on manual inspection and screening, which can be time-consuming, subjective, and lead to inaccurate and inefficient detection results. To address this issue, this study collected and processed pictures of defects in banded board edges, accurately labeling them into six categories, including open glue, shortage, chipping, uneven trimming, glue line, and banding indentation, using the LabelImg software. The YOLOv7 network architecture was then leveraged to train the network on the accurately processed surface defect dataset of banded edge panels. Subsequently, samples of panels produced through the manufacturing process were analyzed and tested, with the training outcomes effectively visualized and analyzed using the TensorFlow tool, resulting in a mean average accuracy (mAP) rate of 74.8% and an average detection rate of 57.63 FPS. The test findings revealed that the YOLOv7 target detection network successfully identified defects during the production of banded edge panel parts, and thus, improved the accuracy and efficiency of detecting banded edge panel defects. Therefore, this method is feasible and holds immense potential for practical application, as indicated by its capability to immensely enhance the accuracy and efficiency of detecting banded edge panel defects.