Workpiece Recognition under Poor Lighting Conditions using a Two-Stage Mask R-CNN Approach

With the advent of Industry 4.0, smart manufacturing has imposed higher demands on automated inspection. Traditional metal workpiece recognition methods often rely on manual inspection or basic image processing techniques, which are easily affected by lighting variations and reflections, leading to low recognition accuracy. This paper explores the application of deep learning, specifically the Mask R-CNN model, for workpiece recognition. To address the limitations of conventional recognition methods in complex environments, we propose a hybrid approach that combines image processing techniques with a two-stage Mask R-CNN framework. The original Mask R-CNN model is fine-tuned and improved into a two-stage architecture, which is then integrated with image processing to overcome challenges caused by complex lighting conditions. This method aims to accurately identify workpieces and determine their correct positions on gantry equipment, thereby fulfilling the requirements of automated measurement on production lines. Experimental results show that our method performs well in terms of precision and recall, offering a novel solution for efficient and accurate workpiece recognition. Furthermore, the proposed system can determine whether the workpiece is within the working area, whether the distance between workpieces meets safety thresholds, and the rotation angle of each workpiece, thus providing a comprehensive solution for automated measurement in smart manufacturing.