基于改进YOLOX-S的火灾检测方法

Fire detection method based on improved YOLOX-S

针对传统火灾检测算法存在检测精度不高,检测速度慢等问题,该算法对当前目标检测领域检测效果较好的YOLOX-S算法改进,提出适合于火灾检测这一特殊领域的检测算法YOLOX-IMP.在YOLOX-S算法基础上,通过对预测头部分多尺度检测的改进,将原来的3尺度改为4尺度检测,对YOLOX-S算法的损失函数改进,在YOLOX-S预测头部分添加SENet注意力,提高火灾检测的检测精度.实验结果表明,改进后算法YOLOX-IMP精确度和mAP值分别提高了2.9%,2.3%,在检测速度未明显下降情况下,该算法火灾检测精度相比YOLOX-S算法有较大提升,证明该算法的可行性.

Aiming at the problems of low detection accuracy and slow detection speed of traditional fire detection algorithm,this paper improves YOLOX-S algorithm that has good detection effect in the field of target detection and proposes a detection algorithm YOLOX-IMP which is extremely suitable for this special field of fire detection.Based on YOLOX-S algorithm,the original 3-scale detection was changed to 4-scale detection by improving the multi-scale detection of the prediction head,and the loss function of YOLOX-S algorithm was improved by adding SENet attention to the prediction head,so as to improve the detection accuracy of fire detection.The results show that the accuracy and mAP value of the improved YOLOX-IMP algorithm are improved by 2.9%and 2.3%respectively.The fire detection accuracy of the improved YOLOX-IMP algorithm is greatly improved compared with the YOLOX-S algorithm under the condition that the detection speed is not significantly decreased,which proves the feasibility of the proposed algorithm.