Research On Flame Detection Algorithm Based On Deep Learning

In unsupervised places(such as warehouses,automated workshops,tunnels,etc.),the prevention of fire is an important measure of ensuring property safety,image-based flame detection method has become a hot research topic.The paper studies the flame detection technology based on deep learning,which has important theoretical significance and practical application value.The paper studies the image pixel characteristics of flame,gives a color criterion rule of flame and extracts the possible flame regions from the single frame image of video,according to the rule;to determine the flame region,a single-frame image flame detection algorithm based on CNN is designed.As the single-frame image flame detection algorithm does not consider the dynamic characteristics of the flame,resulting in a high false detection rate,the paper proposes a continuous frame extraction rule,which can extract continuous multi-frame valid flame region images from video and designs a continuous frame flame detection algorithm based on CNN-LSTM network model,which can determine the dynamic flame of video.More than 6,700 images were used to train the designed single-frame image flame detection algorithm based on CNN and more than 2,600 images were used for testing;the accuracy rate is 91.3% and the recall rate is 98.6%.More than 1,000 images were used to train the designed single-frame image flame detection algorithm based on CNN and more than 400 images were used for testing;the accuracy rate is 94.7% and the recall rate is 96.4%.Favorable detection results are achieved.