Intelligent Monitoring System for Smoke and Fire Detection Based on Improved CenterNet
To address the contradictory nature of computational efficiency and detection accuracy of convolutional neural net-works,this paper proposes an improved CenterNet-based smoke and fire detection algorithm.This paper introduces a lightweight ar-chitecture MobileNetV3 based on CenterNet to replace the original backbone network,which reduces the network size and improves the portability while ensuring the detection accuracy.Further,this paper designs and develops an intelligent monitoring system for smoke and fire detection on offshore platforms based on the idea of microservices.The system can detect the smoke and fire in the video monitoring stream of the offshore oilfield operation site in real time and do alarm processing in time.The improved CenterNet smoke and fire detection algorithm proposed in this paper has an accuracy of 91.5%and a mean average accuracy of 89.8%in the smoke and fire test set,compared to the original CenterNet target detection model,although the accuracy decreases by 0.3%,the detection speed increases by 30 frames/s and the model size decreases by 93 MB.The system is tested on Tesla V100 and can meet the application requirements of real-time processing and detection on offshore platforms.
offshore oil fieldsmoke and fire detectionintelligent monitoring
国家科技期刊平台
NSTL
万方数据
