基于泊松噪声-双边滤波算法的桥梁裂缝修补痕迹图像分割方法

裂缝作为混凝土桥梁的主要病害大量存在，部分裂缝修补后会二次开裂，在病害智能识别中，裂缝修补痕迹易与混凝土剥落等缺陷混淆，因此，裂缝修补痕迹的准确识别不仅是二次裂缝准确识别的基础，也是混凝土桥梁整体病害识别的重要环节。为了获取边缘清晰连续的裂缝修补痕迹，先对裂缝修补痕迹的图像添加泊松噪声，再利用双边滤波对添加的噪声及原有噪声进行平滑。然后用Otsu算法对裂缝修补痕迹进行图像分割，并使用峰值信噪比（PSNR）和结构相似性（SSIM）评价滤波效果，使用运行时间和最大连续可用内存块（LCFB）使用情况评价分割效果。最后以河南省某高速公路桥梁历年定检中的裂缝修补痕迹图像为例，对方法进行实际验证。结果显示：经过泊松噪声-双边滤波算法处理后，裂缝修补痕迹图像PSNR值最高约35.090 1 dB,SSIM值可达约0.880 1，说明添加泊松噪声可改善图像质量并优化双边滤波效果；经过Otsu算法进行图像分割的运行时间比其他方法约短25%～50%,LCFB比其他方法约提高0.25%；经过处理的裂缝修补痕迹图像达到理想预期效果，验证了提出方法的有效性和可行性。

Image segmentation method of bridge crack repair traces based on Poisson-noise and bilateral-filtering algorithm

Large number of cracks, as the main disease, exist in the concrete bridge, and some cracks will be secondary dehisced after maintenance, and the crack repair traces are easily confused with concrete spalling and other defects when identifying disease intelligently, as a result of which identifying the crack repair traces accurately is not only the basis for identification of secondary cracks but also important for identification of the overall disease of concrete bridges. To obtain crack repair traces with continuous edges clearly, Poisson-noise is firstly added to the image of crack repair traces, then bilateral-filtering was adopted to smooth the added and the original noise, the Otsu algorithm was also used to segment the image of crack repair traces. The filtering effect is evaluated using the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), and the segmentation effect is evaluated using the running time and maximum continuous memory block (LCFB) use. The results show that the highest PSNR value of the crack repair trace images processed by the Poisson-noise and bilateral-filtering algorithm is about 35.090 1 dB, and the SSIM value reach about 0.880 1, which shows that adding Poisson-noise improves image quality and optimizes the bilateral filtering effect. The running time of image segmentation by the Otsu algorithm is about 25%-50% shorter than other methods, and meanwhile the LCFB is about 0.25% higher. The processed crack repair trace images achieve the desired effect, which verifies the effectiveness and feasibility of the method proposed.