摘要
火灾场景在视觉检测中呈现出高度的非线性和不确定性,为实现对火灾精准监测,设计基于轻量化深度学习的火灾自动化遥感监测系统。采用遥感传感器集成系统、数据传输与预处理平台、边缘计算与智能识别模块设计硬件结构;采用Ostu算法对最佳阈值进行选取,判断像素点是否属于火焰区域,实现图像前景分割。引入空间注意力机制对CNN模型进行改进,结合卷积操作对其进行特征提取以及训练,输出图像识别结果,判断是否发生火灾情况。实验测试结果表明,采用设计的系统对图像进行火灾识别,火焰区域IoU值较高,具备较为理想的识别效果。
Fire scene presents a high degree of nonlinearity and uncertainty in visual detection.In order to realize accurate monitoring of fire,this paper designs an automatic remote sensing monitoring system for fire based on lightweight deep learning.The hardware structure is designed by using remote sensing sensor integrated system,data transmission and preprocessing platform,edge calculation and intelligent identification module.The Ostu algorithm is used to select the best threshold,judge whether the pixel belongs to the flame area,and realize the image foreground segmentation.The spatial attention mechanism is introduced to improve the CNN model,and the feature extraction and training are combined with convolution operation,and the image recognition results are output to judge whether there is a fire.The experimental test results show that the IoU value of the flame area is high when the designed system is used to identify the fire,and it has an ideal recognition effect.
作者
赵明
张雪燕
周盼
ZHAO Ming;ZHANG Xueyan;ZHOU Pan(Anhui Jiyuan Software Co.,Ltd.,Hefei 230000,China)
出处
《自动化与仪表》
2025年第8期151-154,158,共5页
Automation & Instrumentation
