In Japan, unlike in many other countries, urbanization has progressed while original rural road structures have been retained, leading to distinctive urban sprawl areas with intermingling residential lots and farmland. Currently, much of Japan’s urban areas consist of urban sprawl areas, posing considerable challenges for infrastructure development. However, for such urban sprawl areas in Japan, it is difficult to say that methods have been established to identify their spatial distribution based on quantitative evaluation. Therefore, for this study, we used machine learning to investigate a system that extracts sprawling urban areas from aerial photographs divided into meshes. In the system’s design, we prioritized precision to ensure the reliable detection of urban sprawl areas. Consequently, the accuracy of identifying sprawl areas achieved precision of 0.81, recall of 0.63, and an F-score of 0.71. Examination of the classification results of sprawl areas revealed that most misclassifications occurred near class boundaries. By contrast, areas with particularly high levels of urban sprawl showed few misclassifications.
Hemmi et al. (Fri,) studied this question.