配电柜线号特征提取和识别方法研究

Study of Feature Extraction and Recognition for Distribution Cabinet

配电柜接线状态是影响高速动车组运行性能的重要因素,基于视觉的自动化检测方式可以解决传统的人工检测的低效、主观性强的问题,是工业检测的一个重要的检测方向,由于动车配电柜在光照和空间等方面的恶劣情况,线号数字呈现灰度、尺寸、角度等差异,极不利于线号的识别;以线号正确分割提取为前提,提取分割的数字区域的HOG特征和LBP特征,并利用SVM和ANN两种分类器配合两种特征进行对比试验,因为HOG特征在图像几何和光学的形变会表现出优异的不改变的特性,且SVM分类器克服了后者局部极小的困扰,最终选用SVM分类器配合HOG特征进行配电柜线号识别。

Distribution cabinet is a key equipment of the China Railway High-speed （CRH）. Automatic detection method based on computer vision can solve the subjective problem and low efficiency of traditional manual detection. Bad situation in the light and space and differences in gray seale and angle made recognition more difficult. Based on correct segmentation district , extracting HOG features and LBP features of line number, using SVM and ANN classifiers with the two characteristies to recognize line number. Considering the excellent invariance of HOG features in geometry and optical deformation of the image, and SVM classifier can overcomes local minimum problem, This paper use SVM classifier with HOG characteristics to recognize distribution cabinet line number.