Research on Improved Faster R-CNN in Stacked Artifact Recognition

In order to improve the sorting efficiency of workpieces, improve the sorting quality, reduce the labor intensity of workers and improve the working environment, and at the same time meet the needs of small batch, multi-variety and mixed production, this paper studies the improvement of Faster R-CNN for the field of robotic sorting. Application in stacked workpiece recognition. This paper improves the Faster R-CNN and proposes the Bin-Soft-NMS algorithm based on the Soft-NMS algorithm. The dynamic threshold is used to improve the low recall rate of the model and the low score of the adjacent stacked workpiece bounding box caused by the inaccurate threshold. the problem of inhibition. A stacked workpiece dataset was made by using four kinds of workpieces, and the model was trained on the dataset to complete the experiment. The experimental results show that the average detection accuracy of the improved model in the stacked workpiece scene is 86.26%, and the improvement in mild, moderate and severe scenarios is improved respectively. increased by 1.0%, 3.2% and 4.825%, with an average increase of 3%.

I. Introduction

In industrial production, the workpieces produced by casting and other processes are generally stacked in the hopper or tray, and then the subsequent finishing is carried out through sorting and sorting. Conventional manual sorting methods have the disadvantages of poor working environment, high labor intensity, repeated labor, fatigue and sorting errors. In the automatic production line, specially designed sorting equipment often has the disadvantages of high noise, high failure rate and poor adaptability. Based on the research of Faster R-CNN model, the work intensity of multi material and batch sorting is improved to meet the needs of workers, and the work intensity of sorting is improved at the same time.