BOSC: A toolbox for aerial imagery mapping

TL;DR

The paper addresses the challenge of efficient and accurate labeling of aerial imagery across domains such as agriculture, forestry, and urban planning. It presents BOSC, a client-server toolbox with a backend for heavy computations and a frontend for interactive annotation. The methods comprise a Segment Anything-inspired Segmentation Engine, a ConvNeXt-based unsupervised Classification Engine with optional labeled inputs, and an OpenCV-driven Mapping Engine using three-point affine alignment. BOSC's reported versatility and mapping capabilities aim to accelerate decision-ready insights for drone- and satellite-derived data.

Abstract

Accurate and efficient label of aerial images is essential for informed decision-making and resource allocation, whether in identifying crop types or delineating land-use patterns. The development of a comprehensive toolbox for manipulating and annotating aerial imagery represents a significant leap forward in remote sensing and spatial analysis. In this report, we introduce BOSC, a toolbox that enables researchers and practitioners to extract actionable insights with unprecedented accuracy and efficiency, addressing a critical need in today's abundance of drone and satellite resources. For more information or to explore BOSC, please visit our repository.

Figures (4)

• Figure 1: Client-Server architecture of BOSC.
• Figure 2: Example of crown-tree classification: a) Raw input image. b) Automatic segmentation output. c) Manually fine-tuned segmentation output. d) Default classification. e) Automatic classification output (no labeled object provided). f) Manually fine-tuned classification output.
• Figure 3: Example of mapping between an input image (left) and its corresponding location on a third-party map (center). Both images display three selected points used for the mapping. The final outcome (right) shows the new layer containing the labeld segmentation image overlaid on the interactive map.
• Figure 4: From top to bottom, the first pair of images illustrates an example where BOSC is employed to map only pine trees. This scenario is useful for forest inventory, particularly in estimating the carbon fixation coefficient (net oxygen production), which is tree-dependent. The second example depicts irrigated agriculture in the desert (1 km diameter) in Saudi Arabia. By using the segmentation mask, the status of the crops can be monitored and farming practices optimized. Finally, the third example demonstrates a scenario where multiple classes are mapped, showcasing BOSC's capability to handle diverse scenarios such as semi-urban or urban environments.