改进SSD模型的路面病害图像检测系统

为解决公路路面病害图像特征不突出、检测精度低等问题，提出一种基于改进SSD模型的路面病害检测系统。利用梯度下降Sobel算子优化SSD模型中图像特征提取的卷积网络层，突出路面病害图像特征；通过改进SSD模型实现横向裂缝、纵向裂缝、块状裂缝、路面凹陷以及其它类路面的病害图像检测；结合Jetson-Nano板载化系统以及基于GO语言的Tensorflow框架实现路面病害检测及分类。实验结果表明，系统路面病害分类准确度为91.28%，比未改进的SSD模型识别准确度提高7.36%，证明该优化模型有效。

Pavement Distress Image Detection System Based on Improved SSD Model

A pavement distress detection system based on a modified SSD model is raised in the paper to solve the problems of not-prominent road image features and low accuracy of detection. Firstly， gradient descent Sobel operator is used to modify the convolution neural network extract image features in the SSD model to highlight features of pavement distress pictures. The improved SSD model is used to realize the detection of transverse cracks， longitudinal cracks， block cracks， road depressions and other types of roads. Then the detection and classification of powement distrers is realised by employing Jetson-Nano onboard system and Tensorflow framework based on GO language. The experimental data including collected pavement distress pictures and that generated by the adversarial generation network. Results show that the classifying accuracy of our system is 91.28% which is 7.36% higher than that using the unmodified SSD model， which proves that the proposed optimization model has practicality.