基于深度学习的指针式仪表识别算法研究

Research on Pointer Instrument Recognition Algorithm Based on Deep Learning

目前指针式仪表数据仍然基于模板匹配等传统算法,在信噪比低的情况下识别精度低,因此采用深度学习进行指针的分级定位与识别。一级识别采用SSD算法进行仪表区域定位,计算仪表倾斜角度并修正。二级指针定位使用多方位SSD算法识别指针转动角度并转换量程。使用自建的仪表数据集进行网络训练,指针式仪表检测精度达到87%;作为拓展方向,数字式仪表检测精度达88%。实验结果表明,该算法稳定性及准确度均高于传统的指针式仪表识别算法。

At present,the identification of pointer instruments is still based on traditional algorithms such as template matching,and the identification accuracy is low in the case of strong noise.Therefore,deep learning is used to carry out hierarchical positioning and recognition of pointers.The first level identification uses SSD algorithm to locate the instrument area,calculate the instrument inclination angle and correct it.A multi-directional SSD algorithm is used to identify the rotation angle of the secondary pointer and convert the range.The self-built meter dataset is used for network training,and the detection accuracy of the pointer meter reaches 87%.As an extension direction,the detection accuracy of digital instruments reaches 88%.Experimental results show that the stability and accuracy of the algorithm are higher than those of the traditional pointer meter recognition algorithm.