中文摘要：

针对路面裂缝图像识别结果容易存在孤立噪声和断续边缘的情况，提出了基于像素一裂缝子块双层连通性检测的图像自动识别算法，主要有4个部分：（1）基于自适应灰度拉伸的图像增强算法；（2）基于自适应大津法和八方向Sobel梯度信息的组合分割算法；（3）基于连通性检测的二值图像去噪算法；（4）32×32裂缝子块识别和优化连接算法。然后，对5张3056×2048的路面破损图片进行裂缝识别，结果显示，该算法从像素和裂缝子块这2个层次进行连通性增强处理，可获得完整而连续的裂缝图像。最后，针对10张512×512的路面破损图片，对全局OTSU分割、八方向Sobel检测、Canny检测和本文算法进行测试，各算法综合性能指标Fl值依次为62．46％、23．84％、10．45％和88．30％，准确率依次为83．45％，27．82％，17．83％和86．60％，召回率依次为56．89％，21．83％，8．89％和90．68％，体现了本文算法的优越性。

英文摘要：

Aiming at the fact that image recognition results of pavement cracking tend to have isolated noises and broken edges, an automatic image recognition algorithm based on connectivity checking of pixel and cracking subimage levels is proposed, which has four main parts： （1） image enhancement algorithm based on self-adaptive grayscale stretch; （2） combination image segmentation algorithm based on self-adaptive OTSU and 8-direction Sobel gradient information; （3） binary image denoising algorithm based on connectivity checking; （4） recognition and optimal connecting algorithm of 32 × 32 cracking subimages. Then, the cracking recognition is conducted on five 3 056 × 2 048 pavement damage images, which shows that the cracking recognition results keep better integrity and continuity because of connectivity enhancement on pixel and subimage levels. At last, the performance tests are carried out on ten 512 × 512 pavement damage images by global OTSU segmentation, 8-direction Sobel detection, Canny detection and the proposedalgorithm, from which the values of comprehensive performance index F1 come out as 62. 46%, 23.84%, 10. 45% and 88.30%, the precisions are 83.45%, 27.82%, 17. 83 % and 86.60%, and recalls are 56. 89%, 21.83%, 8.89% and 90. 68%, for each of the algorithms sequentially, which indicated a outstanding performance of our algorithm.