一种改进Faster RCNN的工件检测算法

Improved faster RCNN workpiece detection algorithm

针对在工业自动化生产过程中,光线不佳,工件尺寸较小等外在因素导致的多种工件检测精度不高以及特征提取困难的问题,提出一种改进更快速区域卷积网络(faster region with convolution neural networks,Faster RCNN)的工件检测算法。在原有网络基础上,结合自动色彩均衡算法增加图像预处理模块,改善光照不均匀问题,获得高质量图像。此外,通过增加锚点个数并修改其尺寸优化网络模型,提高网络的拟合能力。实验结果表明,该算法对多种工件的平均检测精度提高了3.6%,符合工业自动化场景要求。

In order to solve the problems of low-precision detection on small-sized workpieces and the difficulty to extract multiple workpiece features due to external factors such as poor light during the industrial automation production process,a workpiece detection algorithm based on a faster regional convolution network(Faster RCNN)is proposed.The original network,an automatic colour equalization algorithm is combined and added to the image preprocessing module to improve the unbalanced lighting and to achieve the high-quality image.In addition,the basic network model is modified by increasing the number of anchors to improve the fitting ability of the network.Experimental results show that compared with the original algorithm,the improved algorithm can increase the average detection accuracy by 3.6%,which meets the requirements of industrial automation scenarios.