A Data-Driven RetinaNet Model for Small Object Detection in Aerial Images

2025-09-02

AI Summary

Small-object detection in aerial imagery suffers from low accuracy, high annotation costs, and data scarcity. To address these challenges, we propose DDR-Net—a data-driven small-object detection framework built upon RetinaNet. Its key contributions are: (1) a data-driven multi-level feature adaptive selection mechanism that dynamically prioritizes discriminative features for small objects; (2) a learnable anchor parameter estimation module that replaces hand-crafted anchors with optimized, scale-aware proposals; and (3) a small-object-oriented dynamic positive/negative sampling strategy to mitigate class imbalance and improve localization precision. DDR-Net enables end-to-end training with limited annotated data, substantially reducing reliance on large-scale labeled datasets. Evaluated on multiple aerial bird datasets, DDR-Net achieves 3.2–5.8 percentage points higher mAP than RetinaNet and state-of-the-art methods, with particularly pronounced gains for objects smaller than 32×32 pixels. The framework demonstrates strong practical applicability in environmental monitoring, urban planning, and public safety.

Technology Category

• Computer Vision: Object Detection & Categorization
• Machine Learning: Multi-instance/Multi-view Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

In the realm of aerial imaging, the ability to detect small objects is pivotal for a myriad of applications, encompassing environmental surveillance, urban design, and crisis management. Leveraging RetinaNet, this work unveils DDR-Net: a data-driven, deep-learning model devised to enhance the detection of diminutive objects. DDR-Net introduces novel, data-driven techniques to autonomously ascertain optimal feature maps and anchor estimations, cultivating a tailored and proficient training process while maintaining precision. Additionally, this paper presents an innovative sampling technique to bolster model efficacy under limited data training constraints. The model's enhanced detection capabilities support critical applications including wildlife and habitat monitoring, traffic flow optimization, and public safety improvements through accurate identification of small objects like vehicles and pedestrians. DDR-Net significantly reduces the cost and time required for data collection and training, offering efficient performance even with limited data. Empirical assessments over assorted aerial avian imagery datasets demonstrate that DDR-Net markedly surpasses RetinaNet and alternative contemporary models. These innovations advance current aerial image analysis technologies and promise wide-ranging impacts across multiple sectors including agriculture, security, and archaeology.

Problem

• Enhancing small object detection in aerial images
• Optimizing feature maps and anchor estimations autonomously
• Improving model efficacy with limited training data

Innovation

• Data-driven RetinaNet model for small object detection
• Autonomous optimal feature map and anchor estimation
• Innovative sampling technique for limited data training