基于时空域深度神经网络的野火视频烟雾检测

针对目前的烟雾检测算法主要基于单一特征或烟雾的多个动静态特征的融合导致检测精度低的问题,提出一种使用卷积神经网络和循环神经网络组合的视频烟雾检测框架来捕获烟雾在空间域和时间域中的特征信息。利用空间流网络部分对运动区域自动提取特征后进行初步的空域的判别;在将空域判断为有烟的基础上进一步通过时间流网络和循环神经网络部分累积一组连续帧之间的运动信息以区分烟雾和非烟雾区域。与现有的使用深度卷积神经网络模型进行对比实验,实验结果表明,该方法具有较高的分类检测准确率。在多个视频场景中进行测试,验证了该算法的有效性。

WILDFIRE VIDEO SMOKE DETECTION BASED ON SPATIO-TEMPORAL DEEP NEURAL NETWORK

Current detection methods are mainly based on the fusion of multiple dynamic and static features of a single feature of smoke, resulting in low detection accuracy. To solve this problem, this paper proposed a video smoke detection framework using a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to detect the feature information of smoke in spatial-temporal domain. We used the part of spatial flow network to identify the initial spatial domain after automatically extracting the features from moving region. Then, on the basis of judging the spatial domain as smoky, we further accumulated a set of motion information between consecutive frames to distinguish smoke and non-smoke regions through time flow network and cyclic neural network. A comparison experiment was carried out with the existing deep convolution neural network model. The experimental results show that the detection results of the proposed method are higher than those of the CNNs model which only extracts spatial features. Experiments in several video scenes demonstrate the effectiveness of the proposed method, and the video smoke detection capability is improved.