基于嵌入式AI的森林火灾图像检测系统设计  

Design of forest fire image detection system based on embedded AI

作　　者：潘骁 王勋[2] 苏冬胜[1] 封李红 王军 Pan Xiao;Wang Xun;Su Dongsheng;Feng Lihong;Wang Jun(School of Industry&Art-Design,Guangxi ECO-Engineering Vocational and Technology College,Liuzhou 545004,China;School of Electronic Engineering,Guangxi Normal University,Guilin 541004,China)

机构地区：[1]广西生态工程职业技术学院,广西柳州545004 [2]广西师范大学,广西桂林541004

出　　处：《无线互联科技》2023年第16期62-66,共5页Wireless Internet Technology

基　　金：2023年度广西教育厅高校中青年教师科研基础能力提升项目;项目名称:基于嵌入式人工智能的森林火灾图像识别与检测研究;项目编号:2023KY1263。

摘　　要：为解决当前森林火灾检测存在的不便和成本高等问题,文章利用堪智K210嵌入式AI微控制器,设计了一种森林火灾烟雾识别系统。该系统利用已有的森林火灾烟雾和火焰图像数据集,采用YOLOv2-Tiny卷积神经网络,经过MaixHub在线Tensorflow深度学习开源框架得到训练模型文件,并将其部署到K210微控制器上。该系统通过OV2640摄像头拍摄图像,一旦图像包含烟雾及火焰形态,可在显示屏上进行识别。测试结果表明,该系统识别的FPS约为13~15帧,识别检测烟雾和火焰准确率分别可达91%和90%。In order to solve the problems of inconvenience and high cost in current forest fire detection,a forest fire smoke recognition system was designed using the embedded AI microcontroller of Kendryte K210.Utilize the existing forest fire smoke image data set,use YOLOv2-Tiny convolutional neural network,obtain the training model file through MaixHub online Tensorflow in-depth learning open source framework,and deploy it to the K210 microcontroller.Take the image through the OV2640 camera.Once the image contains the smog and flame form,it can be identified on the display screen.The test results show that the FPS recognized by the system is about 13~15 frames,and the accuracy of identifying and detecting smoke and flame can reach 91%and 90%respectively.

关 键 词：K210 YOLOv2-Tiny 火灾检测 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TP315[自动化与计算机技术—控制科学与工程]