摘要
野外烟雾具有稀疏性,在整幅图像中所占面积小,扩散缓慢。传统烟雾检测算法由人工提取的特征不一定是最合适的烟雾特征,从而导致烟雾检测存在错检或误检。为此,设计了基于卷积神经网络的野外烟雾检测算法。通过改进的VGG16网络以及搭建的卷积神经网络conv-10进行烟雾检测。VGG16过滤器尺寸大小为3×3,步长为1。conv-10通过对LeNet增加相应的层数进行烟雾检测。实验证明,conv-10网络具有较高的准确率,达到94.7%,时间仅需要1656s。改进的VGG16网络准确率也较高,但是比conv-10网络速度慢,时间需要10450s。
Traditional smoke detection algorithm detects smoke according to the features extracted manually.Because the features extracted manually are not necessarily the most suitable features for smoke,there would be false detection or false detection in smoke detection.Field smoke is sparse,occupies a small area in the whole image and diffuses slowly.In this paper,a field smoke detection algorithm based on convolution neural network is designed.Smoke detection was carried out by improved VGG16 network and conv-10 convolution neural network.The size of VGG16 filters is 3*3 and the step size is 1.Conv-10 detects smoke by adding layers to LeNet.Experiments show that the conv-10 network has a high accuracy rate of 94.7%and only needs 1656 seconds.The network accuracy of the improved VGG16 is relatively high,but compared with the conv-10 network,the speed is slower and the time is 10450s.
作者
张欣欣
ZHANG Xin-xin(College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266590,China)
出处
《软件导刊》
2020年第2期118-121,共4页
Software Guide
