Efficient Flame Recognition Method Based on a Deep Convolutional Neural Network and Image Processing

Abstract

This study proposes an efficient method for implementing continuous flame image tracking and recognition in home environments. This method involves the application of a modified deep convolutional neural network (CNN) in combination with color space conversion, image filtering and morphology, background subtraction, corner detection, and region of interest techniques. Unlike previous recognition methods that only considered flame features, a deep learning architecture was applied to raise our method's accuracy, allowing for the establishment of a modified deep CNN that is more efficient. The test results indicated that the method was able to achieve a flame image recognition rate of 90% and higher. Thus, it is hoped that the method will enable people to perform the real-time tracking and recognition of flame images, such that they can be alerted to the presence of a flame before it starts to expand and threaten the safety of personnel and property.