Aiming at the problems of lack of domestic cement pavement disease detection data sets, small scale, single scene, and insufficient generalization ability of deep learning algorithms in complex scenes, a pavement disease detection algorithm based on improved YOLOv5 is proposed. A cement pavement disease detection dataset containing 11 862 images was collected and constructed, covering 3 most common disease types in 9 scenarios. The prior anchor frame of model training is obtained by combining the K-Means clustering algorithm measured by IoU and genetic algorithm. In the feature enhancement stage, a lightweight upsampling module (CARAFE) is introduced to reduce the information loss in the feature recombination process. A multi-dimensional collaborative attention module (MCA) considering channel, height and width dimensions is introduced to enhance the discrimination of multi-scale disease features. The experimental results show that under the premise of maintaining fast inference speed, the average of F1 score and average precision (mAP) of the proposed algorithm reach 75.5% and 81.6%, respectively, which are better than the five mainstream target detection algorithms. The example analysis shows that the pavement disease detection algorithm based on improved YOLOv5 can meet the actual needs of large-scale cement pavement disease intelligent detection and damage condition evaluation.
为构建具有复杂场景和显著样本差异性的数据集,依托辽宁工程技术大学交通时空大数据研究中心(institute of spatiotemporal transportation date,ISTD),采用辽宁省多年份路面图像数据库。通过专家筛选,从该数据库中挑选出11 862张水泥路面病害图像,构建ISTD-PDD3数据集。数据集图像具有不同的清晰度、光照强度、路面干燥程度,以及包含各种干扰物的情况。此处干扰物包括伸缩缝、防滑槽、阴影、污渍和路面杂物等元素,其目的是提高模型的鲁棒性。病害类型包含常见的水泥裂缝(SC)、破碎板(PSB)和坑洞(KD)。样本图像如图1所示。使用LabelImg进行精准的标注,并保存为Pascal VOC的数据格式。新数据集按8∶1∶1的比例划分训练集、验证集和测试集。无病害的负样本不进行标注处理。样本图像见图1,标注样例见图2。
in complex scene
1.2 数据集对比分析
变异系数(coefficient of variation,CV)可反映不同数据集样本总体的离散程度,选择图像像素灰度的平均值、标准差和信息熵(information entropy,IE)3个度量指标的变异系数分析不同数据集图像场景的复杂度[20]。图像像素灰度平均值可定量反映图像的平均亮度;图像像素灰度的标准差可反映图像像素灰度相对于灰度均值的离散程度,也可反映图像对比度的强弱;IE从信息论的角度衡量图像中灰度分布的聚集特征所包含的平均信息量。
具体而言,对输入的特征向量,首先将其传递给多维协同注意力模块中的3个分支。在分支1中, X 沿H轴被逆时针旋转90°,得到。然后,为构建高度维度和通道维度之间的交互,对挤压变换得到特征图,再通过激励变换捕获通道维度C和空间维度W之间的交互关系,得到W方向上的特征权重。接下来,在W维度上,通过对的矩阵转换和sigmoid函数生成特定的注意力权重;将与逐元素相乘得到具有空间维度W与通道维度C交互关系的特征映射矩阵;最后,将沿轴顺时针旋转90°,得到与原始输入相同形状的特征图。
在分支2中,输入特征向量 X 沿W轴被逆时针旋转90°,得到。之后,类似于分支1中的操作,实现空间维度与通道维度交互关系的建模,最终生成一个原始输入相同形状的特征。
分支3包含通道注意力机制,主要负责从通道域的角度赋予图像不同位置不同的权重,以获得更重要的特征信息。首先,通过恒等映射(identity mapping)生成与 X 相同的特征图。然后,将依次输入到挤压及激励模块中,获得通道特征权重;接着对特征图 X 进行赋值,获得特征映射图。最后,对不同维度的注意力权重做平均聚合,并重新校准三个分支的所有输出,即可推导出更加细化的病害信息特征图,其表达式为
YANBanfu, XUGuanya, LUANJian,et al.Pavement distress detection based on faster R-CNN and morphological operations[J].China Journal of Highway and Transport,2021,34(9):181-193.
[4]
CHENGH D, SHIX J, GLAZIERC.Real-time image thresholding based on sample space reduction and interpolation approach[J].Journal of Computing in Civil Engineering,2003,17(4):264-272.
[5]
AYENU-PRAHA, ATTOH-OKINEN.Evaluating pavement cracks with bidimensional empirical mode decomposition[J].EURASIP Journal on Advances in Signal Processing,2008,2008(1):861701.
[6]
HEY Q, QIUH X, WANGJ,et al.Studying of road crack image detection method based on the mathematical morphology[C]//2011 4th International Congress on Image and Signal Processing. October 15-17,2011,Shanghai,China.IEEE,2011:967-969.
[7]
AMHAZR, CHAMBONS, IDIERJ,et al.Automatic crack detection on two-dimensional pavement images:an algorithm based on minimal path selection[J].IEEE Transactions on Intelligent Transportation Systems,2016,17(10):2718-2729.
[8]
ZHANGD J, LIQ Q, CHENY,et al.An efficient and reliable coarse-to-fine approach for asphalt pavement crack detection[J].Image and Vision Computing,2017,57:130-146.
[9]
LIN N, HOUX D, YANGX Y,et al.Automation recognition of pavement surface distress based on support vector machine[C]//2009 Second International Conference on Intelligent Networks and Intelligent Systems.November 1-3,2009,Tianjian,China.IEEE,2009:346-349.
[10]
HOANGN D.An artificial intelligence method for asphalt pavement pothole detection using least squares support vector machine and neural network with steerable filter-based feature extraction[J].Advances in Civil Engineering,2018(1):7419058.
LIZhangwei, HUAnshun, WANGXiaofei.Survey of vision-based object detection methods[J].Computer Engineering and Applications,2020,56(8):1-9.
[13]
DUZ Y, YUANJ, XIAOF P,et al.Application of image technology on pavement distress detection:a review[J].Measurement,2021,184:109900.
[14]
RAJESHWARIP, ABHISHEKP, VINODP S,et al.Object detection:an overview[J].International Journal of Trend in Scientific Research and Development,2019,3(3):1663-1665.
[15]
LIUY C, LIUF, LIUW,et al.Pavement distress detection using street view images captured via action camera[J].IEEE Transactions on Intelligent Transportation Systems,2024,25(1):738-747.
[16]
LIJ Q, ZHAOX F, LIH W.Method for detecting road pavement damage based on deep learning[C]//Health Monitoring of Structural and Biological Systems XIII. April 4-7,2019,Denver,Colorado,United States. Bellingham.Washingtion:SPIE,2019:517-526.
LUOHui, JIAChen, LIJian.Road surface disease detection algorithm based on improved YOLOv4[J].Laser & Optoelectronics Progress,2021,58(14):336-344.
[19]
MANDALV, MUSSAHA R, ADU-GYAMFIY.Deep learning frameworks for pavement distress classification: a comparative analysis[C]//2020 IEEE International Conference on Big Data.December 10-13,2020.Atlanta,GA,USA.IEEE,2020:5577-5583.
[20]
GAOM X, WANGX, ZHUS L,et al.Detection and segmentation of cement concrete pavement pothole based on image processing technology[J]. Mathematical Problems in Engineering,2020,2020:1360832.
[21]
SADATIS, CHAGAS BRITO DA SILVAL E, WUNSCHD C I I,et al.Artificial intelligence to investigate modulus of elasticity of recycled aggregate concrete[J].ACI Materials Journal,2019,116(1):51-62.
HUIBing, GUOMu, WANGZhou,et al.Multi-dimensional index detection of potholes based on 3D laser technology[J].Journal of Tongji University(Natural Science),2018,46(1):60-66.
GUOXiaoying, LIWenshu, QIANYuhua,et al.Computational evaluation methods of visual complexity perception for images[J].Acta Electronica Sinica,2020,48(4):819-826.
[26]
SUB Y, ZHANGH, WUZ H,et al.FSRDD:an efficient few-shot detector for rare city road damage detection[J].IEEE Transactions on Intelligent Transportation Systems,2022,23(2):24379-24388.
[27]
MAJIDIFARDH, JINP, ADU-GYAMFIY,et al.Pavement image datasets:a new benchmark dataset to classify and densify pavement distresses[J].Transportation Research Record: Journal of the Transportation Research Board,2020,2674(2): 328-339.
[28]
REDMONJ, FARHADIA.YOLO9000:better, faster, stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6517-6525.
[29]
WANGJ Q, CHENK, XUR,et al.CARAFE:content-aware ReAssembly of FEatures[C]//2019 IEEE/CVF International Conference on Computer Vision.October 27- November 2,2019,Seoul,Korea.IEEE,2019:3007-3016.
[30]
CAOY, XUJ R, LINS,et al.GCNet:non-local networks meet squeeze-excitation networks and beyond[C]//2019 IEEE/CVF International Conference on Computer Vision Workshop. October 27-28, 2019, Seoul, Korea. IEEE, 2019: 1971-1980.
[31]
YAOH, LIUY H, LIX,et al.A detection method for pavement cracks combining object detection and attention mechanism[J].IEEE Transactions on Intelligent Transportation Systems,2022,23(11): 22179-22189.
[32]
YUY, ZHANGY, CHENGZ Y,et al.MCA: multidimensional collaborative attention in deep convolutional neural networks for image recognition[J]. Engineering Applications of Artificial Intelligence,2020, 126: 107079.
[33]
RENS Q, HEK M, GIRSHICKR,et al.Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[34]
LIUW, ANGUELOVD, ERHAND,et al.SSD:single shot MultiBox detector[C]//2016 European Conference on Computer Vision, October 8-16,2016,Amsterdam,Netherlands.Berlin:Springer International Publishing,2016:21-37.
[35]
DUANK W, BAIS, XIEL X,et al.CenterNet:keypoint triplets for object detection[C]//2019 IEEE/CVF International Conference on Computer Vision. October 27-November 2,2019,Seoul,Korea. IEEE,2019:6568-6577.
[36]
WANGC Y, BOCHKOVSKIYA, LIAOH M.YOLOv7:trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 17-24,2023,Vancouver,BC,Canada.IEEE,2023:7464-7475.
XUPeng, ZHUXuan, YAODing,et al.Review on intelligent detection and decision-making of asphalt pavement maintenance[J].Journal of Central South University (Science and Technology),2021,52(7): 2099-2117.