Unified Object Detector With Dynamic Switch Mechanism Based on Deep Learning Advances: A Paradigm-Aware Architecture for Adaptive Detection

Abstract

Object detection research has evolved through different architectural paradigms, including two-stage region-based models, one-stage efficiency-focused detectors, and more recent transformer methods, where each performs well in a specific domain but none is optimal overall. This fragmentation requires static architectural decisions by developers and often leads to suboptimal performance when application requirements change. In this work, we address these limitations through two complementary and novel contributions. First, we conduct a systematic bibliometric analysis of the object detection literature from 2014–2025 to derive an empirical paradigm–domain relationship matrix, enabling the identification of the most influential detection paradigms together with their dominant application domains. This provides, for the first time, a data-driven mapping between detection paradigms and domain-specific requirements. Second, we propose a novel Unified Object Detector with Dynamic Switch Mechanism (UOD-DSM), which operationalizes the insights obtained from the bibliometric analysis into an adaptive detection architecture. The UOD-DSM integrates multiple detection paradigms within a unified framework and employs a context-aware switching mechanism to dynamically select the most suitable paradigm at runtime according to operational constraints and input characteristics. Experimental validation demonstrates that the proposed framework offers improved flexibility and robust performance across diverse applications, including real-time video analysis and precision-demanding medical imaging, while outperforming conventional static detectors in cross-domain evaluation. These results demonstrate the potential of the proposed approach to advance object detection toward fully adaptive vision systems capable of aligning architectural strategies with varying operational contexts.