工业场景下基于深度学习的散乱堆叠工件识别

Recognition of Scattered Stacked Workpieces Based on Deep Learning in Industrial Scenarios

针对工业机器人分拣场景下的散乱堆叠工件目标识别率不高、实时性较差的问题,采用检测速度较快且性能较好的深度学习YOLO v3算法对工业场景下的法兰盘和三通管进行目标识别,使用Darknet53网络,采用Kmeans聚类得到先验框的尺寸,并使用logistic对目标进行预测.通过KinectV2相机采集目标RGB数据制作数据集,完成样本的训练.试验结果表明,在工业场景下YOLO v3算法相较于传统算法HOG+SVM对散乱堆叠的目标识别结果准确率更高、识别速度更快,并且具有良好的鲁棒性.

In view of the problem that the target recognition rate of scattered stacked workpieces in industrial robot sorting scenarios is not high,and the real-time performance is poor,the deep learning YOLO v3 algorithm with faster detection speed and better performance is used for flanges and tees in industrial scenes.For target recognition,Darknet-53 network and K-means clustering are applied to obtain the size of the prior frame,and logistic is used to predict the target.Then target RGB data is got by using KinectV2,for the purpose of making a data set,and completing the training of the samples.The experimental results show that compared with traditional algorithm HOG+SVM,the YOLO v3 algorithm is more accurate and faster in recognition,and better in robustness in the recognition of scattered objects in industrial scenarios.