LAUSR

Runway heading is a crucial runway attribute that closely affects aircraft' takeoff and landing safely while having a large number of missing in existing runway thematic databases, limiting the evaluation and analysis globally. This article proposes an automated runway heading extraction method, considering various runway surface materials and spatial structure differences encountered in wide-area detection, promoting a quick and reliable broad investigation. First, multiscale detection is performed based on the mask regional convolutional neural network runway detection model to locate airport runways, using the mask nonmaximum suppression mechanism algorithm to distinguish runways with highly overlapping candidate results. Second, abnormality judgment and mask selection according to semantic constraints were applied to the identified runway area to reserve high-confidence masks to accurately extract the runway heading. We tested our framework results on 1483 runways in the United States with complex runway materials and structures combination. The runway detection recall was 99.12%, and 98.9% of the runway heading errors were limited to 2°, showing that the overall frame for extracting the true runway heading is reliable. In addition, we complement the true headings of 892 runways in OurAirports, promoting research on the completion, verification, and updating of runway semantics in related databases.