Technical Content
Remote Sensing Applications for Road Condition Monitoring by Means of Multispectral Satellite Imagery
Road networks play a crucial role in socio-economical status of nations. Road conditions such as pavement aging and pavement damages affect rideability, safety and user costs. Maintaining high quality for road pavements is a crucial task for highway agencies. To that end, conditions of roads should be regularly assessed to grasp useful information about roads through time. Conventionally, road condition assessment has been done by visiting and inspecting roads. Inspectors use different indexes such as pavement condition index (PCI), International roughness index (IRI), and structural index to determine condition of roads. However, due to hazards related to these labor and time intensive in-situ visits, this method has made many problems. To overcome this issue, remote sensing that had been utilized in various fields, showed potentials to be used in pavement engineering by the mean of monitoring techniques.
To monitor road quality, either as a supplement or as an alternative to the ground-based methods, the application of remote sensing has been investigated by several researchers in last two decades. Remote sensing methods use electromagnetics radiations in different regions of electromagnetic spectrum to obtain information about the target object. In pavement engineering, one of the most common means of remote sensing is multispectral imagery. Multispectral images comprise a limited set of electromagnetic channels or bands, possessing wide bandwidth that facilitates the capture of signals from very far distances. This characteristic permits the deployment of multiple multispectral sensors into the orbit for the acquisition of multispectral images. When compared to other land categories, monitoring of roads via multispectral satellite images demands a significantly high spatial resolution. With recent enhanced spatial resolution of the multispectral sensors, these platforms have attracted significant attention among pavement researchers interested in multispectral imagery. However, despite the use of very high-resolution images, models were unable to achieve very high accuracy and the mixed pixel issue remains a as the main challenge in using multispectral satellite imagery in road pavement monitoring. The integration of machine learning techniques with remote sensing has emerged as a promising approach for monitoring pavement conditions
By leveraging machine learning algorithms to analyze this data, it becomes possible to automate the detection, classification, and tracking of pavement conditions over time, as well as predict future deterioration. In the first phase of the project, the preliminary study was done on both concrete and asphalt pavement sections using both multispectral and hyperspectral images to track changes of spectral profiles of concrete and asphalt pavements over time. It was seen that overall, concrete pavements have higher reflectance values in all the spectrum than asphalt pavements. Figure 2 shows the reflectance values (Red band) of the studied asphalt pavement section obtained from a satellite platform from 2010-2022. As seen in this figure, maintenance work can be detected from satellite imagery over time. Also, the result of hyperspectral scanning showed that there will be changes in the spectral properties of asphalt pavements because of aging and degradation by traffic passes.
In the second phase, our group has developed a model to utilize multispectral satellite images and ML algorithms to detect the type of pavements, and the conditions of asphalt pavements in a very large highway facility. Using historical satellite data, the dataset was generated with over 5.8 million datapoints and the result of classification showed an accuracy 74% in detecting road type and conditions.
