基于空洞卷积与动态多核融合池化的裂缝检测  

Crack detection based on dilated convolution and dynamic multi-kernel fusion pooling module

作　　者：杨秋媛 李宁[1] 石林[1] 庄丽华[1] 徐守坤[1] YANG Qiu-yuan;LI Ning;SHI Lin;ZHUANG Li-hua;XU Shou-kun(School of Computer Science and Artificial Intelligence&Aliyun School of Big Data&School of Software,Changzhou University,Changzhou 213164,China)

机构地区：[1]常州大学计算机与人工智能学院阿里云大数据学院软件学院,江苏常州213164

出　　处：《计算机工程与设计》2022年第12期3529-3537,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(61906021);常州市城市大数据分析与应用技术重点实验基金项目(CM20193007);江苏省油气储运技术重点实验室基金项目(CDYQCY201901)。

摘　　要：针对现有裂缝检测方法在裂缝拓扑结构复杂、前后景像素不协调情况下产生的检测精度低的问题,提出一种基于空洞卷积和动态多核融合池化的裂缝检测方法。采用编、解码结构,编码器阶段插入空洞卷积,能够更大限度保留裂缝的细节与结构信息;解码器阶段引入动态多核融合池化模块,以更高效获得不同尺寸的裂缝信息。在自制数据集与公共数据集CRACK500上分别进行实验,并与其它算法进行比较,实验结果表明,该算法能更加精细化分割出细小的裂缝,有效提高裂缝检测精度。Aiming at the problems of high detection difficulty and low detection accuracy caused by the existing crack detection methods in the cases of complex crack topology and incongruity of foreground and background pixels,a crack detection method based on dilated convolution and dynamic multi-kernel fusion pooling module was proposed.Encoder-decoder structure was adopted.In the encoder stage,dilated convolution was inserted to retain the details and structure information of the crack image.In the decoder stage,the multi-kernel pooling module was introduced to obtain the crack information of different sizes more efficiently,which effectively detected the tiny cracks in the image.Experiments were carried out on self-made datasets and public datasets CRACK500,and the results were compared with that of other algorithms.Experimental results show that the algorithm can more finely segment small cracks and effectively improve the accuracy of crack detection.

关 键 词：图像分割 裂缝检测 编解码网络结构 空洞卷积 动态多核融合池化模块 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]