Abstract
Arson detection plays a critical role in protecting lives and property in high-risk environments such as airports, industrial zones, and other public areas. Recent advances in deep learning (DL), particularly YOLOv10, have demonstrated strong potential in real-time object detection. However, the model's performance is highly dependent on effective hyperparameter optimization to maintain a balance between accuracy and computational efficiency. This study proposes a hybrid GWO-BBOA optimization algorithm that combines the global search capability of Grey Wolf Optimization with the fine-tuning strength of the Brown Bear Optimization Algorithm to optimize YOLOv10 for arson detection. The model was evaluated using an augmented dataset of 2,182 annotated images. Experimental results show that the proposed GWO-BBOA approach outperforms traditional optimization methods, achieving a recall of 0.620. This indicates its enhanced ability to detect true arson events. Moreover, the hybrid algorithm effectively balances exploration and exploitation while reducing the number of required iterations. Future work will focus on expanding the dataset, implementing adaptive optimization strategies, and integrating the model into real-time surveillance systems. Overall, this work highlights the value of hybrid metaheuristic approaches in enhancing DL models for safety-critical applications.