基于YOLOv5的加油站火灾视频图像智能识别

Intelligent Recognition of Fire Video Image in Gas Station Based on YOLOv5

针对当前加油站现场监控预警过程中可能存在的早期明火识别反应慢等问题,通过现场模拟实验及网络获取相结合构建了10万余例火灾图像数据集,改进了YOLOv5s神经网络结构,研发了适用于石化加油站等场景的早期火焰目标检测模型。实验结果表明,改进后的模型在识别精度、召回率和平均识别准确率等指标方面均有提升,随机抽取加油站火灾事故图像做效果测试,达到100%的识别精度和96%的召回率。在此基础上,构建加油站早期火灾智能监测平台,为突发性火情下的应急响应提供有效的预警支持。

In view of the possible problems such as slow response of early open flame identification in the current on-site monitoring and early warning process of gas stations,more than 100000 cases of fire image dataset were constructed through on-site simulation experiments and network acquisition,the YOLOv5s neural network structure was improved,and an early flame target detection model suitable for petrochemi-cal gas stations and other scenes was developed.The experimental results showed that the improved model had improved in recognition accuracy,recall rate and average recognition accuracy,etc.Random sampling of fire accident images of gas stations for effect testing can achieve 100%recognition accuracy and 96%re-call rate.On this basis,the intelligent monitoring platform of early fire of gas station was constructed to provide effective early warning support for emergency fire response under sudden fire.