基于改进 Resnet18网络的火灾图像识别
- 分享到:
摘要：
针对传统卷积神经网络进行火灾图像识别时,准确率不高、特征难以提取、网络的平移不变性较弱等问题,对 Resnet18网络进行改进,使其具有更高的性能和准确性。首先,在 Resnet18网络的卷积层前插入空间变换网络(spatialtransform networks,STN)。对于连续多个相同大小的卷积层,只在第一个卷积层前添加STN,共添加5个,并且在全连接层后添加dropout层防止过拟合。然后,使用迁移学习(transferlearning,TL)的方法对火灾进行分类识别。实验结果表明,改进后的Resnet18网络准确率、召回率、F1值和AUC值等各项指标性能优于Resnet18网络和其他深度学习识别算法,能够对火灾图像进行快速、准确地识别。
In view ofthe problems such as low accuracy, difficultfeature extraction and weak translation invariance of the network duringfireimagerecognition bytraditionalconvolutionalneuralnetwork, thispaperimproved Resnet18network to make ithave higherperformance and accuracy. First, the spatialtransformation network(STN) is inserted in frontofthe con- volution layeroftheResnet18network. Formultiple convolution layersofthe samesizein a row, onlytheSTN isadded before the firstconvolution layer, a totalof five are added, and the dropoutlayer is added after the fully connected layer to prevent overfitting. Then, the transferlearning(TL) method is used to classify and identify fires. Experimental results show thatthe improved Resnet18 network accuracy rate, recall rate, F1 value and AUC value are superior to Resnet18 network and other deep learning recognition algorithms, and can quickly and accurately identify fire images.
作者：
陈跨越,王保云
Chen Kuayue,Wang Baoyuna
机构地区：
云南师范大学数学学院；云南省现代分析数学及应用重点实验室
引用本文：
陈跨越,王保云.基于改进Resnet18网络的火灾图像识别[J].betway官方app 学报(自然科学版),2024,52(4): 101-110.
Chen Kuayue,Wang Baoyun. Fire image recognition based on improved Resnet18 network[J].Journal of Henan NormalUniversity(Natural Science Edition),2024,52(4):101-110.DOI:10.16366/j. cnki.1000-2367.2023.05.19.0001.
基金：
国家自然科学基金
关键词：
火灾检测 ;卷积神经网络 ;空间变换网络 ;Resnet18;HSI色彩模型 ;迁移学习
fire detection; convolutionalneural network; spatial transformation network; Resnet18; HSI color model; transferlearning
分类号：
TP391. 41;TP183;X928. 7
