Innovative Flame and Smoke Detection Framework Leveraging Odconvbs (Omni-Dimensional Dynamic Convolution with Batch Normalization and Silu Activation) - Yolov5s (You Only Look Once)

Timely and accurate identification of flames and smoke is essential in mitigating fire-related hazards. Traditional detection systems, often reliant on sensors, are constrained by limited range and sensitivity to environmental interference. This research introduces an enhanced real-time detection model based on YOLOv5s, incorporating ODConvBS to improve attention-aware feature learning. Additional modifications, including Gnconv integration and Shuffle Attention at the neck, strengthen spatial feature representation and fusion. The use of the SIOU loss function accelerates convergence and refines localization. The proposed system achieves significant improvements in both detection accuracy and responsiveness, making it suitable for real-time fire surveillance in safety-critical applications.