Adaptive Transformer-Based Multi-Object Detection Using Hybrid Metaheuristic Optimization

Detecting multiple objects in surveillance scenes is challenging due to scale variations, occlusions, background clutter, and changing illumination, which limit the performance of existing object detection models. To address these challenges, this paper proposes an Adaptive Transformer-Based Multi-Object Detection Framework optimized using a hybrid Fire Hawk Optimization (FHO) and Arithmetic Optimization Algorithm (AOA). The proposed framework integrates Contrast Limited Adaptive Histogram Equalization (CLAHE)-based image enhancement, Gaussian noise normalization, a Vision Transformer with multi-scale feature fusion, and channel-spatial attention to extract discriminative features and achieve precise object localization. The hybrid optimization strategy employs FHO for effective global exploration and AOA for fine-grained parameter refinement during the exploitation stage, resulting in improved convergence and detection performance. Experimental evaluation on the VisDrone dataset demonstrates superior performance over state-of-the-art CNN- and transformer-based baseline models, achieving higher Precision, Recall, F1-Score, mAP@0.5, and Intersection over Union (IoU). These results indicate that the proposed framework provides a reliable and efficient solution for real-time surveillance applications, including intelligent traffic monitoring, public safety, smart city infrastructure, autonomous surveillance, and security management.