This paper presents an advanced approach to enhance emergency vehicle detection by integrating acoustic and visual information using deep learning techniques. The authors developed an attention-based temporal spectrum network (ATSN) for ambulance siren sound detection and an enhanced Multi-Level Spatial Fusion YOLO (MLSF-YOLO) architecture for visual detection. These models were combined using a stacking ensemble learning technique to create a robust framework for emergency vehicle detection. The ATSN uses attention mechanisms to focus on specific time intervals and frequency ranges in acoustic spectrograms, while the MLSF-YOLO incorporates multi-level spatial fusion to capture deep-level semantic information. The experimental results show that the proposed system achieves a misdetection rate of only 3.81% and an accuracy of 96.19% when applied to visual data containing emergency vehicles, demonstrating significant progress in real-world applications. The study also addresses limitations in existing systems, such as computational inefficiency and the need for diverse datasets, and provides practical applications for improving road safety and emergency response systems.This paper presents an advanced approach to enhance emergency vehicle detection by integrating acoustic and visual information using deep learning techniques. The authors developed an attention-based temporal spectrum network (ATSN) for ambulance siren sound detection and an enhanced Multi-Level Spatial Fusion YOLO (MLSF-YOLO) architecture for visual detection. These models were combined using a stacking ensemble learning technique to create a robust framework for emergency vehicle detection. The ATSN uses attention mechanisms to focus on specific time intervals and frequency ranges in acoustic spectrograms, while the MLSF-YOLO incorporates multi-level spatial fusion to capture deep-level semantic information. The experimental results show that the proposed system achieves a misdetection rate of only 3.81% and an accuracy of 96.19% when applied to visual data containing emergency vehicles, demonstrating significant progress in real-world applications. The study also addresses limitations in existing systems, such as computational inefficiency and the need for diverse datasets, and provides practical applications for improving road safety and emergency response systems.