基于LBP和稀疏表示的林火烟雾图像识别研究

Forest Fire Smoke Recognition Based on Local Binary Patterns and Sparse Representation

为了实现森林火灾的智能识别,提出一种基于局部二值模式和稀疏表示的林火烟雾自动识别方法。选取森林火灾烟雾视频和干扰视频,经运动区域检测提取疑似林火烟雾图像样本。首先对疑似林火烟雾样本图像采用不同的LBP算子进行纹理特征提取,然后选取350幅林火烟雾样本构建林火烟雾特征字典,另外选取343幅林火烟雾图像样本和331幅干扰图像样本进行测试,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稀疏表示系数和重构误差,最后根据经验阈值进行分类识别。结果表明,LBP特征提取结合稀疏表示方法可以有效地实现林火烟雾的自动识别,识别率可达92.88%,为林火烟雾的模式识别提供了一种有效的解决方案。

In order to achieve intelligent recognition of forest fires,an automatic recognition method of forest fire smoke based on Local Binary Patterns （LBP） and sparse representation was proposed.A lot of forest fire smoke video and disturbance video were collected.Extracting image samples suspected of forest fire smoke by motion region detection on the collected videos.Firstly,the texture features of forest fire smoke sample images were extracted by using different Local Binary Pattern （LBP） operator.Then select 350 samples of forest fire smoke images to build a forest fire smoke dictionary for the sparse representation.The other 343 forest fire smoke image samples and 331 samples of disturbance image were treated as the testing samples.For each test sample,its sparse representation coefficients and reconstruction error on the training dictionary were calculated by using l1-minimization.Finally,classification results were obtained based on experience thresholds of the reconstruction error.The experimental results show that LBP feature extraction combined with sparse representation can effectively achieve automatic recognition of forest fire smoke.The overall recognition rate can reach 92.88％.The research can provide an effective solution for pattern recognition on forest fire smoke.