Research on Automatic Identification of Diagonal Keyholes Based on Railway Container Side Crane Robotic Arm
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
维普
万方数据
针对集装箱侧面吊机械臂对角件锁孔的识别问题进行研究,结合深度学习技术提出改进的Cen-terNet目标检测算法.首先,选取DLA(Deep Layer Aggregation)为骨干网络,将关键点高斯热图由圆形改进为椭圆形,提高算法对锁孔的识别准确率;其次,在骨干网络 DLA 中添加 RFB(Receptive Field Block)模块以增大算法对锁孔的感受野;最后,将ReLU(Rectified Linear Uint)激活函数改进为PReLU(Parametric Rectified Linear Uint)以解决神经元"死亡"问题.在集装箱锁孔数据集上完成对比试验,试验结果表明改进后的CenterNet目标检测算法对集装箱锁孔的识别准确率为99.8%,处理速度为19.61 帧/s,可满足电气化铁路场站的集装箱装卸需求.
In order to study the recognition problem of the keyhole of the corner part of the container side crane manipulator,an improved CenterNet object detection algorithm was proposed by combining deep learning tech-nology.Firstly,DLA(Deep Layer Aggregation)was selected as the backbone network,and the Gaussian heat map of key points was improved from a circle to an ellipse to improve the accuracy of the algorithm for keyhole recognition.Secondly,the RFB(Receptive Field Block)module was added to the backbone network DLA to increase the receptive field of the algorithm to the keyhole.Finally,the ReLU(Rectified Linear Unit)activa-tion function was improved to PReLU(Parametric Rectified Linear Unit)to solve the problem of neuronal"death".The results show that the improved CenterNet object detection algorithm has an accuracy of 99.8%and a processing speed of 19.61 FPS for container keyhole recognition,which can meet the needs of container loading and unloading in electrified railway stations.
side craneelectrified railway stationsdeep learningGaussian heatmap
维普
万方数据
