An Improved Fire Detection Algorithm Based on YOLOv5

Early warning of fire is crucial for improving survival rates during fire incidents. Fire and smoke detection is an extremely important safety technology that plays a vital role in timely and accurately detecting fires in the surrounding environment to ensure the safety of lives and production environments. To address the challenge of detecting fires in complex scenarios, we propose a YOLO-based algorithm, utilizing YOLOv5 as its foundation, which offers high detection accuracy while remaining lightweight. Our algorithm incorporates the C2f-ODConv module into the network, allowing for dynamic adjustment of the convolution kernel's shape and size based on the input data's features during the convolution process. This adaptability ensures better compatibility with different input data types. Additionally, we introduce the Decoupled head in the detection head section, replacing the original detection head. This modification reduces interference between the classification head and the detection head of the model, thereby enhancing the overall detection accuracy of the model. Experimental results demonstrate that our proposed algorithm achieves a detection accuracy of 94.7% and a recall rate of 93.2%. These impressive results ensure accurate Decoupled head fire detection even in complex situations.