基于改进型残差网络烟雾图像识别

Smoke Image Recognition Based on Improved Residual Network

当前在深度学习上对烟雾图像和视频识别较少,目前存在的问题是烟雾视频图像第一帧识别率低、覆盖范围小、自适应较差的情况。基于卷积神经网络,改变Res Net(残差网络)结构,实现精确的烟雾区域检测。在实验中经过5000张不同烟雾图像的数据集学习,实验结果准确地识别了烟雾图片,为大范围的火灾烟雾报警提供了一种有效方案。

At present,smoke image and video recognition is less in depth learning.The current problem is that the first frame of smoke video image has low recognition rate,small coverage and poor self-adaptation.The residual network(ResNet)structure by this algorithm was changed and the accurate smoke area detection(was achieved).In the experiment,after 5000 data sets of different smoke images,the experimental results accurately identified the smoke picture,providing an effective solution for a wide range of fire smoke alarms.