基于YOLOv3和光流的船舶火灾检测

Ship Fire Detection Based on YOLOv3 and Optical Flow

为了提高船舶火灾检测的检测效率,在基于视频的方式下,提出将原始帧和光流融合的深度学习检测方法。首先,利用模拟试验和部分公开数据集建立火灾视频数据集;其次,对火灾视频进行处理,提取原始帧和对应的光流;然后,使用像素级融合的方法融合原始帧和光流,充分利用视频的静态信息和动态信息;最后,使用YOLOv3算法,并利用迁移学习方法,实现火灾检测模型的训练。实验结果表明,所用算法能够更有效地识别烟雾和火焰,显著地减少了火灾视频中的误检情况,进行抽帧检测时能够满足实时检测的要求。

In order to improve the detection efficiency of the ship fire detection,based on the video,a deep learning detection method combining original frames and optical flow is proposed.Firstly,simulation experiments and part of the public datasets are used to establish a fire video dataset.Secondly,the fire videos are processed to extract the original frames and the corresponding op-tical flow.Then,the pixel-level fusion method is used to fuse the original frames and optical flow to make full use of the static infor-mation and dynamic information of the videos.Finally,the YOLOv3 algorithm and transfer learning method are used to achieve the training of the fire detection model.The experimental results show that the algorithm used can more effectively identify smoke and flame,significantly reduce the false detection of fire videos,and can meet the requirements of real-time detection when performing frame detection.