Texture-aware object detection and virtual coordinates positioning system by deep learning

Due to the complex and directional texture of the wooden board surface, the traditional object detection methods are prone to misidentification due to texture confusion and object occlusions. Hence, a Texture-Aware Object Detection and Virtual Coordinates Positioning System by Deep Learning (TAODVCP) with the bias covariance matrix module to the Neck layer of YOLOv8 was proposed to distinguish wood textures and reduce the detection error caused by texture similarity. The experimental results showed that the accuracy in TAODVCP could be up to 89% and higher than the accuracy in the traditional object detection methods, such as CenterNet, in scenarios involving overlapping wooden boards and changes of wooden board texture. In addition, TAODVCP proposed in this paper has also been successfully deployed on a Raspberry Pi 5 embedded system to verify the feasibility of real-time reasoning of edge devices in smart manufacturing and automated production systems.