Highway Construction Progress Assessment Based on DeeplabV3+

In order to conduct comprehensive surveys and quantitative assessments of highway construction projects accurately and rapidly, this paper proposes a new method for highway construction progress assessment that combines remote sensing imagery and deep learning. Firstly, high-resolution remote sensing images of highways that are under construction, are obtained using low-altitude unmanned aerial vehicle technology. A deep learning training dataset is then created. DeepLabV3+deep learning network is employed to train the classification model for construction progress assessment. Comparative experiments based on U-net and the maximum likelihood estimate are carried out. The experiment demonstrates that the classification results based on DeeplabV3+ are superior to those of U-net and maximum likelihood estimate, providing necessary data and technical support for the assessment of highway construction progress. Additionally, we evaluated the robustness of this model in terms of image resolution, image noise, and application scenarios. We also drew some conclusions that are helpful for the application of deep learning in engineering.