Applied software systems

Research Article

AST: Joint detection of fences and their shadows on orthophotos of low-rise residential areas

Igor Victorovich Vinokurov

	Financial University under the Government of the Russian Federation, Moscow, Russia
	igvinokurov@fa.ru

Abstract. This paper presents a method for joint detection of fences and their shadows in orthophotos of low-rise residential areas (nadir geometry), built upon the Adaptive Shadow Transformer (AST) architecture. The method relies on three key components:
(1) learnable separate displacement fields for fences and shadows with tailored regularization;
(2) a cross-attention mechanism that incorporates solar azimuth and elevation, as well as fence typology;
(3) a shadow fragment connectivity loss that enables the model to perceive terrain-broken shadow segments as a unified object.
The evaluation is performed on the collected FSAD dataset, comprising 650 manually annotated images and 2,000 synthetic scenes. The proposed approach demonstrates superiority over state-of-the-art detectors (YOLO26, RT-DETR, Deformable DETR, and DINOv2+DETR), providing an absolute mAP@0.5 gain of up to 5.9 percentage points on scenes with complex terrain. (Linked article texts in English and in Russian).

Keywords: aerial imagery, object detection, fences, shadows, vision transformers, deformable attention, cadastral mapping

MSC-2020 68T20; 68T07, 68T45

Acknowledgments: The author expresses gratitude to the students of the Financial University under the Government of the Russian Federation who participated in the annotation of the FSAD dataset. The work was carried out with the support of the Ministry of Digital Development of the Kaluga Oblast within the framework of the project on the development of artificial intelligence technologies.

For citation: Igor V. Vinokurov. AST: Joint detection of fences and their shadows on orthophotos of low-rise residential areas. Program Systems: Theory and Applications, 2026, 17:4, pp. 57–136. (in Engl. In Russ.). https://psta.psiras.ru/2026/4_57-136.

Full text of bilingual article (PDF): https://psta.psiras.ru/read/psta2026_4_57-136.pdf (Clicking on the flag in the header switches the page language).

The article was submitted 10.08.2026; approved after reviewing 21.09.2026; accepted for publication 22.09.2026; published online 16.10.2026.