基于WLD-LPQ特征的心盘螺栓故障图像检测算法
Research on fault recognition algorithm of center plate bolts based on WLD-LPQ features

针对货车心盘螺栓故障图像,提出一种基于韦伯局部描述符(Weber Local Descriptor , WLD)特征和局部相位量化(Local Phase Quantization, LPQ)特征融合的故障图像检测算法。该算法分为螺栓定位与识别2个步骤,采用间接定位法逐步定位感兴趣区域(Region of Interest, ROI),通过最大类间方差法(Otsu)和改进的Canny算法与图像投影结合定位螺栓部位;利用WLD特征和LPQ特征分别提取螺栓纹理,对其归一化,使用PCA降维特征向量,串行联合为WLD-LPQ特征,通过支持向量机SVM自动检测识别。研究结果表明:本文算法能快速准确定位,空间域与频率域结合的WLD-LPQ特征提高了检测识别率,平均检测识别率达到98%以上。
A method based on the combination of WLD and LPQ features was proposed which aimed fault image of center plate bolts of freight train. The algorithm was divided into two steps: bolt positioning and recognition. Firstly, the positioning of the ROI was investigated by the indirect location method. The Otsu Method and the improved Canny algorithm combined with image projection were used to position the bolt. Secondly, the WLD feature and LPQ feature were chosen to extract the bolt texture separately which were normalized next. Then this paper used the PCA dimensionality vector. Finally, the WLD feature and LPQ feature was proposed to form the WLD-LPQ feature which was combined with SVM for automatic recognition and detection. Test results show that this method can quickly locate the center plate bolts. The recognition rate is high and the average fault recognition rate is over 98%