基于BEMD和KELM的路面病害检测算法

Pavement defects detection algorithm based on BEMD and KELM

受外界环境以及道路材料本身影响,路面会出现破损。尽管裂缝是路面破损的首要表现形式,但是其他类型病害仍然占重要比重。针对传统路面病害检测算法对常见线性裂缝分类准确度较高但对车辙、松散等复杂病害识别效果一般且适应性较差的问题,提出一种基于二维经验模态分解(BEMD)与核极限学习机(KELM)相结合的复杂路面病害识别方法。该方法首先采用二维经验模态分解对路面病害图像进行筛分,然后结合主成分分析法对分解后得到的固有模态分量进行降维,最后将上述得到的新特征输入到核极限学习机中进行训练。实验结果表明该算法对复杂病害有较高的识别率,其中松散病害识别率为95.6%,车辙病害识别率为92.1%,坑洼病害识别率为96.9%,网状裂缝识别率为97.3%,与传统脉冲耦合卷积神经网络相比,该算法提高了约9.85%。

Pavements are subject to defects due to the external environment as well as the road material itself.Although cracks are the primary defects of pavement defects,other shapes of defects still play an important role.In allusion to the problem that the traditional pavement defects detection algorithm has a high accuracy in classifying common linear cracks but a general effect and poor adaptability in identifying complex defects such as rutting and loosening,a complex pavement defects identification method based on two⁃dimensional empirical mode decomposition and kernel limit learning machine is proposed.In this method,two⁃dimensional empirical modal decomposition is used to screen the pavement defect images,the dimensionality reduction of the intrinsic modal components obtained after sieving is conducted by means of the principal component analysis,and the new feature vectors obtained above are input to the kernel limit learning machine for training.The experimental results show that this algorithm has a high recognition rate for complex defects,including 95.6%for loose defects,92.1%for rut defects,96.9%for potholes,and 97.3%for network cracks,which is nearly 9.85%higher than the traditional pulse⁃coupled convolutional neural network.