A review of intelligent recognition methods for small aerial targets based on deep learning

Abstract

With the rapid development of the low-altitude economy, the accurate recognition of low, slow and small (LSS) aerial targets (pixel ratio < 1%) has become an urgent demand in airspace supervision, national defense security and other related fields. Traditional recognition methods are difficult to meet the practical application requirements due to the inherent characteristics of small aerial targets, such as tiny scale, weak signals and severe complex background interference. Deep learning has become the core technology to break through the bottleneck of small aerial target recognition by virtue of its powerful automatic feature extraction and complex pattern recognition capabilities. This paper systematically reviews the deep learning-driven intelligent recognition technology for small aerial targets. First, the research foundation, core characteristics and technical difficulties of small aerial target recognition are clarified, including the scarcity of target features, poor environmental anti-interference ability, and the inherent contradiction between real-time performance and recognition accuracy. Then, the core algorithm paths are elaborated from four aspects: anchor-based, anchor-free, single-modal adaptive, and few-shot lightweight deep learning algorithms, with a detailed analysis of the improvement strategies, performance indicators and applicable scenarios of typical algorithms. Furthermore, the key performance optimization strategies (data augmentation, network structure optimization, etc.) and experimental evaluation system (common datasets, core evaluation indicators, algorithm performance comparison) are summarized. Finally, the typical application scenarios of this technology in low-altitude economy supervision, national defense air defense, urban emergency rescue and aerospace security are expounded, and the future development trend is prospected. This review provides a comprehensive reference for the theoretical research and engineering application of deep learning-based intelligent recognition of small aerial targets.