Aggregation segregation generative adversarial network (AG-GAN) facilitated multi-scale segregation detection in asphalt pavement paving stage

Abstract

Pavement segregation usually cause engineering quality problems such as early damage of asphalt pavement. Therefore, it is of great significance to carry out segregation detection in the construction process. However, existing image detection technology has two problems: 1) the segregation boundary and the index are not clear; 2) the segregation assessment is localized. In order to find useful solutions to these two problems, this study proposed an Aggregation Segregation Generative Adversarial Network (AG-GAN) to generate virtual pavement images with segregation failure. Moreover, a multi-scale fusion method of Recurrent Neural Network (RNN) and the Global Averaging Sliding Window (GASW) algorithm is developed to achieve globalized aggregation evaluations. Based on extensive experiments tests, conclusions obtained that the accuracy rate of re-current neural network for segregation determination of pavement sequence can be up to 84.62%. In addition, automatically assessment of the segregation condition of pavement section can be achieved, which proves the practical application effectiveness of the proposed algorithm in the actual paving process.