针对炼油厂火灾安全隐患早期检测的实际需求,本文提出一种基于改进YOLO(you only look once,你只需看一次)的火灾检测方法.首先,对检测头网络模块进行改进,以提高模型对火灾目标的识别精度;其次,优化骨干网络模块结构,增强特征提取能力;再者,通过改进颈部网络模块,实现不同尺度特征信息的有效融合;最后,设计了一种新的损失函数,以平衡模型的检测精度和实时性能.通过消融实验,验证了提出方法中各改进策略的有效性.结果表明,改进后的YOLO在火焰检测任务上不仅展现了出色的检测精度,而且保持了较高的处理速度,实现了精度和速度之间的最佳平衡.该方法不仅为炼油厂火灾安全隐患的早期检测提供了有效的技术手段,而且具有一定的推广和工程应用价值.
Refinery Fire Image Detection Method Based on Improved YOLO
In response to the practical need for early detection of fire safety hazards in refineries,a fire detection method based on the improved You Only Look Once(YOLO)is proposed.We start with improvement of the detection head network module to enhance the recognition accuracy of the model for fire targets.Then,the structure of backbone network modules is optimized to enhance feature extraction capabilities.Furthermore,by improving the neck network module,effective fusion of feature information at different scales can be achieved.Finally,a new loss function was designed to balance the detection ac-curacy and real-time performance of the model.The effectiveness of each improvement strategy in the proposed method has been verified through ablation experiments.The results indicate that the improved YOLO not only demonstrates excellent de-tection accuracy in flame detection tasks,but also maintains a high processing speed,achieving the optimal balance between accuracy and speed.This study not only provides an effective technical means for early detection of fire safety hazards in re-fineries,but also has certain promotion and engineering application value.
万方数据
维普
