道路裂缝图像识别的算法研究

发布时间：2018-02-26 02:18

本文关键词： 裂缝图像裂缝检测简化脉冲耦合神经网络特征提取　出处：《郑州大学》2017年硕士论文　论文类型：学位论文

【摘要】：我国公路的发展一日千里。道路质量的及时检测在延长道路使用寿命的同时,也可以避免路面病害给行车安全方面带来的隐患。考虑到人工检测方法存在效率低、精度低、危险系数较高等缺陷,路面裂缝的自动检测系统成为道路养护方向的热点研究问题,而裂缝自动检测算法则为自动检测系统的核心内容。依据国内外已有的道路裂缝检测相关算法,本文对裂缝检测算法进行了研究与设计。首先,单纯地从对裂缝图像进行裂缝检测的角度出发,需要对传统的脉冲耦合神经网络(PCNN)模型进行简化改进,这不仅可以降低传统PCNN在模拟过程中的计算复杂度,而且保留了其原有的神经元运行特征,使其可以应用于裂缝图像的目标检测。针对PCNN无法确定裂缝图像的最优检测以及脉冲门限具有非线性因子的问题,提出了一种基于遗传算法(GA)和简化PCNN的裂缝图像检测方法—GA-PCNN。该方法采用改进后的最小误差准则作为遗传算法的适应度函数,并且根据遗传算法具有全局最优解的特点确定简化PCNN中各因子的值,实现了简化PCNN的裂缝图像自动分割。在使用GA-PCNN算法对裂缝图像进行处理后,通过一种形态学的抗噪多结构元素边缘提取算子对其裂缝边缘进行提取,然后使用一种基于生长的连接方法对断裂的裂缝块进行边缘连接。基于MATLAB R2009a平台对本文算法进行实验仿真,通过与不同的检测方法进行比较,以区域对比度、ROC曲线这些客观指标为基准对其性能进行分析。分析结果表明,该方法对裂缝图像检测具有较好的有效性与通用性。最后,对利用上述方法得到的裂缝目标图像,进行裂缝特征信息的提取、分类及计算。经过对检测后的图像设置一系列判定条件,提取图像中连通域信息;同时通过观察裂缝在坐标轴投影所呈现的特点,对目标裂缝进行分类;通过细化的方法提取裂缝骨架,并对目标裂缝的面积及长度、宽度信息进行计算。
[Abstract]:With the rapid development of highway in our country, the timely detection of road quality can not only prolong the service life of the road, but also avoid the hidden danger brought by the road surface diseases to the driving safety. Considering the low efficiency and low precision of the manual detection method, Because of the defects such as high risk coefficient, the automatic detection system of pavement cracks has become a hot research issue in the direction of road maintenance. The automatic crack detection algorithm is the core of the automatic detection system. According to the existing road crack detection algorithms, this paper studies and designs the crack detection algorithm. It is necessary to simplify and improve the traditional pulse coupled neural network (PCNN) model from the point of view of crack detection in fracture images, which can not only reduce the computational complexity of traditional PCNN in the process of simulation. Moreover, it preserves its original neuronal characteristics, which can be applied to target detection of crack image. In view of the problem that PCNN can not determine the optimal detection of crack image and the pulse threshold has nonlinear factor, A crack detection method named -GA-PCNN based on genetic algorithm (GA) and simplified PCNN is proposed. The improved minimum error criterion is used as the fitness function of genetic algorithm. According to the characteristics of global optimal solution of genetic algorithm (GA), the value of each factor in simplified PCNN is determined, and the crack image of simplified PCNN is segmented automatically. After using GA-PCNN algorithm to process the crack image, The edge of crack is extracted by a morphological edge detection operator with anti-noise and multi-structure elements. Then a growth-based connection method is used to connect the fracture blocks. The algorithm is simulated based on the MATLAB R2009a platform, and compared with different detection methods. The performance of the method is analyzed based on the objective index of regional contrast and ROC curve. The analysis results show that the method is effective and universal for crack image detection. Finally, the fracture target image obtained by the above method is analyzed. After setting a series of judgment conditions to detect the image, extract the information of the connected region in the image, and observe the characteristics of the crack projection on the axis, the paper carries out the extraction, classification and calculation of the feature information of the crack, and sets a series of judgment conditions to the detected image. The target crack is classified and the crack skeleton is extracted by thinning method, and the information of the area, length and width of the target crack is calculated.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】