基于改进DeepLabv3+网络的沥青道路裂缝检测方法

Asphalt Road Crack Detection Method Based on Improved DeepLabv3+ Network

摘要:

针对传统的语义分割技术对于沥青道路裂缝的检测存在检测精度低、误差大的问题,提出了一个基于改进DeepLabv3+网络的语义分割方法。该方法在编码器阶段,采用轻量级MobileNetv2取代DeepLabv3+的主干网络Xception,从而减少参数量；在解码器阶段,引入双注意力机制以进一步提高网络的分割精度；使用Dice Loss函数与原始交叉熵损失函数混合,以缓解样本中前景和背景不平衡问题。最后以道路实时检测的数据为对象进行了大量的实验,结果表明,该方法与原始DeepLabv3+相比,平均交并比(mIoU)、平均像素精度(mPA)分别提升了8.98%和17.39%。与其他主流语义分割模型相比,改进后的DeepLabv3+在沥青道路裂缝的检测上也取得了较好的效果。

Abstract:

A semantic segmentation method based on an improved DeepLabv3+network is proposed to address the issues of low detection accuracy and large errors associated with traditional semantic segmentation techniques for detecting asphalt road cracks. In the encoder stage, this method replaces the backbone network Xception of DeepLabv3+ with lightweight MobileNetv2, thereby reducing the number of parameters. In the decoder stage, a dual attention mechanism is incorporated to further improve the segmentation accuracy of the network. The Dice loss function is combined with the original cross-entropy loss function to alleviate the imbalance between foreground and background in the sample. Extensive experiments were conducted on real-time road detection data. The results indicate that, compared with the original DeepLabv3+, the average intersection-to-union ratio (mIoU) and average pixel accuracy (mPA) achieved by the proposed method were higher by 8.98% and 17.39%, respectively. As compared with other mainstream semantic segmentation models, the improved DeepLabv3+ also exhibits good performance in detecting asphalt road cracks.