基于稀疏表示的森林火灾火焰识别研究

为了实现森林火灾的智能识别,提出一种基于稀疏表示的林火火焰自动识别方法.以林火火焰和5类干扰物体为研究对象,每类对象从视频图像中随机选取50帧作为训练样本,150帧作为测试样本.对每幅图像提取疑似火焰区域,求取面积变化率、颜色、纹理和形状特征参数.所有训练样本的特征向量构建训练样本特征字典,对每个测试样本利用l1最小化范数计算其在训练字典上的投影系数,根据最小重构残差进行分类识别.结果表明,稀疏表示方法的识别率可达到93.56％,为林火火焰识别提供了一个有效的解决方案.

Recognition of Forest Fire Flame Based on Sparse Representation

An identification method based on sparse representation was proposed for automatic recognition of forest fire flame. The video images of forest fire flame and five varieties of disturbance were taken as the research objects. For each variety, fifty images were selected randomly as the training samples; one hundred and fifty images were treated as the testing samples. For each sample, the regions with fire-like colors were roughly separated. The area change ratio, colors, textural and shape features of these regions were extracted. Feature vectors of all the training samples made up data dictionary of the sparse representation. The projection of the test image on the dictionary was calculated by using ll-minimization. The classification result could be achieved according to the minimum residual error. The experiment results demonstrated that the overall recognition accuracy of the proposed algorithm was 93.56%. Therefore, the proposed method can provide an effective method for recognition of forest fire flame.