MAVSD: A Multi-Angle View Segmentation Dataset for Detection of Solidago Canadensis L.

Abstract

Recent advancements in computer vision and deep learning have advanced automated vegetation monitoring, creating new opportunities for invasive species management. To this end, we introduce MAVSD (Multi-Angle View Segmentation Dataset), specifically designed for detecting Solidago canadensis L., a globally significant invasive plant. The dataset comprises high-resolution images captured by unmanned aerial vehicles from four angles (30°, 45°, 60°, and 90°), providing comprehensive coverage of plant structures and enabling in-depth understanding from multiple perspectives. MAVSD includes pixel-level semantic segmentation annotations across 13 classes, meticulously categorizing vegetation and environmental elements. Extensive experiments with state-of-the-art segmentation models validate MAVSD’s effectiveness in enhancing invasive species detection and monitoring, with multi-angle training improving mIoU by up to 11% over single-angle baselines. The dataset’s multi-angle, high-resolution characteristics strengthen ecological monitoring capabilities, offering valuable resources for research and environmental protection applications.