Predicting road quality using high resolution satellite imagery: A transfer learning approach

Abstract

Recognizing the importance of road infrastructure to promote human health and economic development, actors around the globe are regularly investing in both new roads and road improvements. However, in many contexts there is a sparsity—or complete lack—of accurate information regarding existing road infrastructure, challenging the effective identification of where investments should be made. Previous literature has focused on overcoming this gap through the use of satellite imagery to detect and map roads. In this piece, we extend this literature by leveraging satellite imagery to estimate road quality and concomitant information about travel speed. We adopt a transfer learning approach in which a convolutional neural network architecture is first trained on data collected in the United States (where data is readily available), and then “fine-tuned” on an independent, smaller dataset collected from Nigeria. We test and compare eight different convolutional neural network architectures using a dataset of 53,686 images of 2,400 kilometers of roads in the United States, in which each road segment is measured as “low”, “middle”, or “high” quality using an open, cellphone-based measuring platform. Using satellite imagery to estimate these classes, we achieve an accuracy of 80.0%, with 99.4% of predictions falling within the actual or an adjacent class. The highest performing base model was applied to a preliminary case study in Nigeria, using a dataset of 1,000 images of paved and unpaved roads. By tailoring our US-model on the basis of this Nigeria-specific data, we were able to achieve an accuracy of 94.0% in predicting the quality of Nigerian roads. A continuous case estimate also showed the ability, on average, to predict road quality to within 0.32 on a 0 to 3 scale (with higher values indicating higher levels of quality).

Semantic Segmentation and Edge Detection—Approach to Road Detection in Very High Resolution Satellite Images

“…To segment the road surface from the remote sensing images, MRENet [26] used an encoder-decoder architecture similar to UNet [3] and a PSP module with four subregions between the encoder and the decoder modules. Brewer et al [27] adopted a transfer learning approach based on CNN architecture trained on data collected in the United States and then fine-tuned data collected from Nigeria. Brewer et al [27] developed an Android application to collect road data and used it as input to test several CNN architectures.…”

“…Brewer et al [27] adopted a transfer learning approach based on CNN architecture trained on data collected in the United States and then fine-tuned data collected from Nigeria. Brewer et al [27] developed an Android application to collect road data and used it as input to test several CNN architectures. The proposed approach achieved an accuracy of 80%, with 99.4% of predictions falling within the actual or adjacent class.…”

Deep learning fusion of satellite and social information to estimate human migratory flows

“…With the growth of convolutional neural network‐based approaches to satellite imagery analysis, studies are now beginning to emerge which seek to quantify explicit attributes about geographic locations—that is, the income of a household (Babenko et al, 2017; Jean et al, 2016; Perez et al, 2017; Tingzon et al, 2019), likelihood of a conflict event (Goodman et al, 2020), population density (Hu et al, 2019; Tiecke et al, 2017), school education outcomes (Runfola et al, 2021), and continuous grades of road quality (Brewer et al, 2021; Cadamuro et al, 2018). Many of these studies have been in response to the critical lack of data on human well‐being in data‐scarce environments (Burke et al, 2021), specifically seeking to improve our ability to capture relationships in impoverished areas (Jean et al, 2016).…”

“…Many of these studies have been in response to the critical lack of data on human well‐being in data‐scarce environments (Burke et al, 2021), specifically seeking to improve our ability to capture relationships in impoverished areas (Jean et al, 2016). Among other contributions, this literature has established the value of transfer learning in overcoming the relatively small‐N of many socioeconomic datasets (Brewer et al, 2021; Goodman et al, 2020; Jean et al, 2016; Runfola et al, 2021).…”

Scaling-Up Deep Learning Predictions of Hydrography From Ifsar Data in Alaska

“…Transfer learning methods apply feature relations (in the form of weights from a trained model) learned from one domain and apply them to another domain, which can greatly reduce training requirements for the new domain (Kimura et al, 2020). For example, a transfer learning process achieved 94 percent accuracy in predicting the quality of roads in Nigeria based on a model originally trained to predict road quality in the United States (Brewer et al, 2021). In our application, a base model trained in one geographic area is transferred to another geographic target area to reduce training requirements for the target area.…”