引入MobileNet的轻量化森林火灾视频监测方法研究

Research on the Lightweight Forest Fire Video Monitoring Method by MobileNet

应用视频火灾检测技术能够有效地提高森林火灾监测和预警能力,在生态保护和社会公共安全方面具有重要意义,为此研究适合森林火灾检测的图像静态、动态特征检测方法。同时,使卷积神经网络YOLO轻量化,引入GAM在准确率、模型尺寸和速率上进行平衡。优化的模型在牺牲准确率1.9%的情况下,参数量降低约80%,准确率在自制森林火灾数据集上达到92.4%。这一种基于颜色-运动-机器学习技术相结合的火灾监测新方法总体轻量精简,对实时火灾监测系统设计具有参考价值。

The application of video fire detection technology can effectively improve the monitoring and early warning capabilities of forest fires,which is of great significance in ecological protection and social public safety.The static and dynamic feature detection methods of images that are more suitable for forest fire detection have been studied.At the same time,it makes the convolutional neural network YOLO lightweight,and GAM is introduced to balance accuracy,model size and speed.The optimized model reduces the number of parameters by approximately 80%at the expense of 1.9%decrease in accuracy,and the accuracy reaches 92.4%on the self-made forest fire dataset.This new fire monitoring method based on the combination of color,motion and machine learning technology is lightweight and streamlined,and has reference value for the design of real-time fire monitoring system.