A video flame detection method based on two-stream convolution neural network

Abstract

The invention relates to a video flame detection method based on a dual-stream convolution neural network. The main process comprises the following steps: S1, preparing data, inputting a training dataset into the dual-stream convolution neural network for training, and obtaining a trained network model; 2, pretreating that video to be measured by cascade motion feature detection and color featuredetection to obtain a suspected flame region; S3, inputting the obtained suspected flame region into the trained two-stream convolution neural network for classification and identification; S4: identifying The region as the flame by the two-stream convolution neural network is output as the final detection result. The invention efficiently processes most of the non-flame regions in the video by extracting the suspected flame regions, and reduces the time complexity. By using two-stream convolution neural network, combining the video flame motion information, and adding the spatio-temporal pyramid pooling layer to provide more robust spatio-temporal characteristics, the detection accuracy is improved.

Description

technical field [0001] The invention relates to the field of video image detection, in particular to a video flame detection method based on a two-stream convolutional network. Background technique [0002] Fire promotes the progress of human beings from weak to strong, and gives them light, security and warmth. However, fire has also brought huge harm and loss to society and life. The occurrence of fire is uncertain, which makes it difficult to prevent and detect flames, especially in complex environments, how to detect fires in time is a thorny problem. The traditional sensor-based flame detection technology has the disadvantages of small detection range, low reliability and slow speed. In recent years, the flame video image detection technology based on computer vision has overcome the main weakness of the traditional flame detection technology, so that the visual flame detection technology can combine a large number of dynamic and static characteristics of the flame, w...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a video flame detection method based on a dual-stream convolutional neural network. Aiming at the problem that the flame detection method based on deep learning does not combine video motion information, by using a dual-stream convolutional neural network, a single frame of video can be detected. At the same time as the static feature recognition of the image, the optical flow sequence of continuous multi-frame images of the video is analyzed, and the two are fused to improve the recognition and detection effect

Images