Generalization ability in extreme scenarios.

Currently, forest fires have become a major fire safety issue. To detect forest fires and optimize the accuracy, a forest fire detection and recognition model based on an improved YOLOv5-ACE algorithm is proposed. In response to the difficulties of small target detection in forest fires, poor adaptability to complex backgrounds, and deployment limitations of edge devices, the CBAM and the ASPP multi-scale feature extraction module are introduced to enhance the ability of target feature capture and small target detection. The algorithm is lightweight by combining the grouped convolution of ShuffleNet v2 and the global dependency capture of ViT, while improving the positioning accuracy and anti-interference ability. Compared with the traditional YOLOv5, the detection accuracy has increased by 11.5%, ultimately reaching 92.3%, and the recall has increased by 6.8% to 91.6%. Through hypothesis testing, all performance improvements have statistical significance (p 0.05). The proposed method can detect forest fires more quickly and accurately, which has good guiding significance for preventing the occurrence of forest fires.