Skip to content
Open access

FEAST-DET: Edge-Aware Multiscale Aggregation and Context-Aligned Small-Object Detection for UAV Imagery

2026 · IEEE Access · Vol 14, pp. 170373-170388 · 0 citations · 37 references

Abstract

Small objects in uncrewed aerial vehicle (UAV) imagery occupy few pixels, vary substantially in scale, cluster densely, and often exhibit degraded boundary and texture cues. These properties lead to missed detections and localization errors for general-purpose detectors. We propose FEAST-DET (Fine-grained Edge-Aware Small-Target Detector), a feature-enhanced framework for small-object detection in UAV imagery. First, the multiscale edge-sensitive aggregation backbone (MESANet) uses MESA blocks that integrate a dual-branch topology, multiscale receptive-field fusion, high-frequency detail enhancement, and spatial attention guidance. This design preserves object contours, textures, and salient regions while reducing detail loss during downsampling. Second, the context-aware cross-scale alignment module (CCAM) combines multibranch scale alignment, global-context modeling, and spatial attention refinement to establish long-range semantic dependencies across feature levels and reduce background interference with weak small-object responses. Third, the small-object prediction network (SPNet) adds a <inline-formula> <tex-math notation="LaTeX">$P_{2}$ </tex-math></inline-formula> prediction branch and GSConv-based Slim-Neck components to retain shallow high-resolution features at modest parameter cost. On VisDrone2019, FEAST-DET achieves 55.5% precision, 46.8% recall, 48.2% <inline-formula> <tex-math notation="LaTeX">$\mathrm {mAP}\text {@}0.50$ </tex-math></inline-formula>, and 29.0% <inline-formula> <tex-math notation="LaTeX">$\mathrm {mAP}\text {@}0.50{:}0.95$ </tex-math></inline-formula>, exceeding YOLO11s by 6.7, 8.6, 9.8, and 6.2 percentage points, respectively. The model uses 10.19 M parameters, only 0.77 M more than the baseline. Quantitative comparisons and qualitative analyses across diverse scenes support the effectiveness of FEAST-DET for UAV small-object detection.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.