计算机与现代化

• 图像处理 • 上一篇    下一篇

复杂背景下基于深度卷积神经网络的森林火灾识别

  

  1. (北京林业大学工学院,北京 100083)
  • 收稿日期:2016-01-06 出版日期:2016-03-17 发布日期:2016-03-17
  • 作者简介:傅天驹(1990-),男,天津人,北京林业大学工学院硕士研究生,研究方向:林火识别和机器人技术; 通信作者:郑嫦娥(1977-),女,山西五台人,副教授,研究方向:林火识别和机器人技术。
  • 基金资助:
    国家自然科学基金资助项目(31200544); 中央高校基本科研业务费专项资金资助项目(YX2013-14); 高等学校博士学科点专项科研基金资助项目(20110014120012)

Forest Fire Recognition Based on Deep Convolutional Neural Network Under Complex Background

  1. (School of Technology, Beijing Forestry University, Beijing 100083, China)
  • Received:2016-01-06 Online:2016-03-17 Published:2016-03-17

摘要: 针对森林火灾的特点,提出并设计一种基于深度学习的森林火灾图像识别方法。通过实验,给出用于复杂背景下森林火灾识别的深度卷积神经网络结构,并对该结构进行训练和测试。并且,针对小样本林火识别存在识别率低的问题,提出一种参数替换方法。结果表明,该方法具备较高的正确率,正确率达到98%。同时网络可自动提取特征,无需对输入图像进行复杂预处理,克服了传统算法许多固有的缺点,将其应用在森林火灾识别领域取得了很好的效果。

关键词: 图像处理, 森林火灾识别, 深度学习, 卷积神经网络

Abstract: According to the characteristics of forest fire, a forest fire image recognition method based on deep learning is proposed and designed. The structure of convolutional neural network (CNN) is given by experiment, which is used in forest fire recognition under the complex background, and it has been trained and tested. A parameters replacement method is presented for low recognition rate existing in small samples forest fire recognition. The results show that the method is of a high accuracy reaching to 98%, it can extract features automatically, the input image doesn’t need to pre-processing, and it overcomes many inherent shortcomings of traditional algorithm. Its application in the field of forest fire recognition achieves good results.

Key words: image processing, forest fire recognition, deep learning, convolutional neural network

中图分类号: