中圖分類號：TP391文獻標識碼：ADOI:10.19358/j.issn.2097-1788.2024.10.008
引用格式：杜世澤，銀皓，豐大軍,等.基于改進YOLOv8的森林火災探測技術研究［J］.網(wǎng)絡安全與數(shù)據(jù)治理，2024，43（10）：49-56,82.

Research on forest fire detection technology based on improved YOLOv8

Abstract： Forest fire detection is a key research direction at present. However, there are a large number of negative sample data in real forest fire scenarios, which seriously affects the performance of target detection. At the same time, edge to edge deployment requires more lightweight models. To address this issue, this article proposes an improved YOLOv8 method, which firstly introduces the EfficientViT module to the backbone network and reduces computational overhead by cascading group attention modules. Then, the CBAM module is introduced into the head network to enhance the features extracted by the backbone network, while suppressing noise and irrelevant information. Finally, for low-quality samples in the dataset, the Wise-IoU loss function is introduced to enhance the training effect of the dataset. The experimental results show that the improved YOLOv8 model achieves a detection accuracy of 79.5% for forest fires, a detection speed of 75 FPS, and a 5.7% reduction in the parameter count of the entire model, which is of great significance for forest fire detection.

Key words : YOLOv8; forest fire detection; image analysis; EfficientViT; attention mechanism