森林小面积火灾烟雾增强识别仿真  

Simulation of Enhanced Recognition of Smoke from Small Forest Fire

作　　者：杨鸿雁 周汝良[1] 王艳霞[1] YANG Hong-yan;ZHOU Ru-liang;WANG Yan-xia(School of Geography and Ecotourism,Southwest Forestry University,Kunming Yunnan 650224,China)

机构地区：[1]西南林业大学地理与生态旅游学院,云南昆明650224

出　　处：《计算机仿真》2023年第1期359-363,共5页Computer Simulation

基　　金：国家自然科学基金地区科学基金项目《基于遥感热辐射增量的早期野火自动识别研究》(编号:C160903,3186030269)。

摘　　要：提出森林小面积火灾烟雾增强识别方法,以实现在火灾初期及时发现并控制,防治林火。上述方法通过构建分形模型提取森林小面积火灾烟雾分形特征,基于傅里叶变换采用二维小波变换分解图像并识别烟雾边缘结构,得到可选择性高的小波系数维度以及低频分量子图和高频分量子图,经小波熵计算后,提取图像纹理特征,将分形特征和纹理特征输入YOLOv3网络模型,利用其内置残差模块具备的目标细粒度检测特点的三个尺度特征图,通过两次筛选,实现森林小面积火灾烟雾增强识别。实验证明:烟雾由于本身特性,高低频相对能量较低以及小波熵数值较高,表现出的纹理性较差,边缘结构不如环境中其它干扰物清晰;所提方法可实现森林小面积火灾烟雾增强识别,且选取1/8尺寸特征图识别小面积火灾烟雾准确率更高。This paper puts forward a enhancement recognition method for smoke from small area forest fire, so as to realize the timely detection and control in the early stage of fire and prevent forest fire. In this method, the fractal features of forest small area fire smoke were extracted by constructing a fractal model. Based on Fourier transform, two-dimensional wavelet transform was used to decompose the image and identify the smoke edge structure. The wavelet coefficient dimension with high selectivity, low-frequency component subgraph and high-frequency component subgraph were obtained. After wavelet entropy calculation, the image texture features were extracted, The fractal features and texture features were input into the yolov3 network model, and the three scale feature maps of the target fine-grained detection features of its built-in residual module were used to realize the enhanced recognition of small-area forest fire smoke through twice screening. Experiments show that due to the characteristics of the smoke itself, the relative energy of high and low frequencies is low and the wavelet entropy value is high, showing poor texture, and the edge structure is not as clear as other interfering objects in the environment. This method can enhance the recognition of small forest fire smoke, and the accuracy of identifying small forest fire smoke by selecting 1/8 size feature map is higher.

关 键 词：分形模型 小波变换 低频分量子图 纹理特征 先验框 小波熵 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]