基于卷积神经网络的林火烟雾识别

Image Identification of Forest Fire Smoke Based on Convolutional Neural Network Algorithm

基于传统的人工神经网络(Artificial Neural Network, ANN)提出了一种算法用来进行森林林火烟雾的图像识别。应用卷积神经网络(Convolutional Neural Network, CNN)结合反向传播法(Backpropagation, BP),选取适当的激励函数,训练神经网络。同时通过适当的池化方法大大提高了算法的效率,从而有效地通过神经网络对目标图像的特征学习,识别出烟雾图像。在对图像识别学习前对图像进行灰度化,并且在对图像进行二值化之后,再进行学习训练,排除了所需识别目标之外图像引入的干扰,从而提高了图像识别准确率。

An improved algorithm is proposed for image identification of forest fire smoke,based on traditional Artificial Neural Network(ANN)algorithm.By applying Convolutional Neural Network(CNN)algorithm,combined with Backpropagation(BP)algorithm,and selecting a proper active function,this algorithm will train the neural network to identify the images.Simultaneously,by a proper pooling method,the effectiveness of the algorithm will be improved significantly and the algorithm will identify the images very accurately.And before learning and identifying the image,the algorithm gets the gray scale image and then the binary image.Finally,it uses the binary image to identify and train,which also improves the accuracy,excluding the interference coming from other part of the image other than the target.The experiment shows the new algorithm possesses good practicability.