Research on Forest Fireworks Detection Algorithm Based on Improved YOLOv8n

In response to the limitations of existing forest fire detection methods regarding accuracy, real-time performance, and adaptability, this paper presents a novel fire detection model by optimizing and enhancing the network architecture of the YOLOv8n algorithm. Firstly, we replace the original C2f with depthwise separable convolution C2f-DSconv to reduce computational load while maintaining feature extraction capabilities. Secondly, we introduce the Swin Transformer to better capture long-range dependencies and global information, thereby improving detection accuracy in distant areas. Furthermore, we incorporate an efficient attention mechanism, SE, enabling the model to focus more precisely on critical features. Lastly, addressing the shortcomings of the traditional CIoU loss function in terms of target localization accuracy and generalization ability, we propose substituting the CIoU loss function with the Wise-IoUV3 loss function. Experimental results on the M4SFWD dataset indicate that the improved model achieves increases of 4.1, 2.6, 3, and 5.1 percentage points in precision, recall, mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50</inf>, and mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50-95</inf>, respectively, compared to the original model. This model not only enhances detection accuracy but also improves adaptability to varying environmental conditions and scene changes, while optimizing computational efficiency for faster and more accurate forest fire monitoring.