火灾探测
计算机科学
人工智能
特征(语言学)
机制(生物学)
遥感
传感器融合
模式识别(心理学)
数据挖掘
融合
目标检测
特征提取
钥匙(锁)
融合机制
计算机视觉
领域(数学)
人工神经网络
机器学习
特征选择
支持向量机
作者
Yuanmao Chen,Yinsheng Chen,Tong Sun,Zedong Ju
出处
期刊:
[SPIE]
日期:2026-05-14
卷期号:: 10-10
摘要
The timely and precise detection of forest fires is paramount for the protection of forest ecosystems.However,the boundary between flames and smoke is often unclear, and they are visually mixed together, especially in the early stages of a fire, when smoke may partially obscure the flames, reducing their visibility. This further increases the risk of false positives and missed detections in fire detection. To address these issues, this study proposes an improved wildfire detection method, YOLOv11n-SDLF, based on the YOLOv11n framework. The method achieves the following innovations: First, the StarNet replaces the original YOLOv11n backbone, reducing the model's burden while improving performance and adaptability. Second, the DySample method is used to replace the nearest-neighbor algorithm, reducing computational complexity and enhancing upsampling efficiency. Third, the LWT module replaces standard convolution, effectively balancing the detection of flames and smoke. Additionally, the FASFFHead detection head enhances the model's ability to detect small objects and complex scenes. Extensive experiments on the M4SFWD dataset show that the improved model achieves an accuracy of 86.3%, a 3% increase over the traditional model. The results demonstrate that, compared to current state-of-the-art object detection algorithms, this method can more accurately detect wildfires.
科研通智能强力驱动
Strongly Powered by AbleSci AI