摘要
研究结合Darknet-53的火灾图像特征提取技术的应用方法,通过大量火灾图像的收集,采用像素点尺度直接计算方式,确定图像数据中的火灾运动目标。在火灾图像的连续帧序列变化下,对火灾图像中具有代表性的特征进行分解,分别按照颜色特征、动态特征进行划分,区分与归类火灾的具体特征向量。选择Darknet-53作为特征提取模型,在模型内对预处理图像进行训练,以Darknet-53的深度神经网络传播模式,针对颜色与动态火灾特征向量建立目标矩阵,实现火灾图像不同目标特征的提取。以不同场景下的火灾图像作为测试数据进行验证可知,所研究的方法可以实现火灾特征的完整分割,并且可以准确地提取到连续帧图像中的火灾特征数据值,具有较高的应用价值。
Image processing and feature extraction technology can effectively solve the problems in using sensors to detect fires.Therefore,the application method of fire image feature extraction technology combined with Darknet-53 is studied.By collecting a large number of fire images and directly calculating the pixel scale,the fire motion targets in the image data are determined.Under the continuous frame sequence changes of fire images,decompose representative features in the fire images,divide them according to color features and dynamic features,and distinguish and classify the specific feature vectors of the fire.Select Darknet-53 as the feature extraction model,train the preprocessed images within the model,and use Darknet-53's deep neural network propagation mode to establish an objective matrix for color and dynamic fire feature vectors,achieving the extraction of different target features from fire images.Using fire images from different scenarios as test data for verification,it can be seen that the method studied can achieve complete segmentation of fire features and accurately extract fire feature data values from continuous frame images,which has high application value.
作者
杨斐
YANG Fei(School of Information and Intelligent Manufacturing,Fuyang Institute of Technology,Fuyang,Anhui 236031)
出处
《武夷学院学报》
2024年第12期6-11,共6页
Journal of Wuyi University
基金
安徽省高等学校科研计划自然科学重点项目(2023AH052417)。
关键词
火灾图像
特征提取
深度学习神经网络
灾情分析
数据处理
fire images
feature extraction
deep learning neural network
disaster analysis
data processing
作者简介
杨斐(1983-),男,回族,副教授,主要从事边缘计算及人工智能方向的研究。
