Flame detection method, flame detection model training method, storage medium and system

Abstract

The invention relates to a flame detection method. The method comprises the following steps: S1, acquiring an image containing flame; S2, providing a first convolutional neural network, inputting theacquired image into the first convolutional neural network to extract a first feature, and calculating a thermodynamic diagram in combination with the first feature; S3, calculating according to the thermodynamic diagram to obtain a flame area thermodynamic mask diagram, and obtaining a suspected flame area based on the flame area thermodynamic mask diagram; S4, providing a second convolutional neural network, and extracting a second feature from the suspected flame region; and S5, connecting the first features with the second features to obtain total output features, and predicting whether flames exist or not, the proportion of the flames in the whole image and the flame hyperactivity based on the total output features. The invention further provides a storage medium. The invention further provides a flame detection model training method. The invention further provides a flame detection model training system.

Description

【Technical field】 [0001] The invention relates to the field of computer vision, in particular to a flame detection method, a flame detection model training method, a storage medium and a system. 【Background technique】 [0002] The early method of computer vision to detect fire is to use artificially extracted features, such as image channel, flame edge, flame physical characteristics, etc., but these solutions are not strong in anti-interference, prone to false negatives, false positives, and image background , High quality requirements. [0003] With the use of deep learning, people began to use the method of deep learning target detection to detect fire images, mostly through the convolutional neural network (CNN, Convolutional Neural Networks) to learn the image features of flames, and then when the monitoring image appears in the same When the flame image features learned by the convolutional neural network, it is to detect the presence of flames in the monitored scene....

AI Technical Summary

Problems solved by technology

[0002] The early method of computer vision to detect fire is to use artificially extracted features, such as image channel, flame edge, flame physical characteristics, etc., but these solutions are not strong in anti-interference, prone to false negatives, false positives, and image background , high quality requirements

[0003] With the use of deep learning, people began to use the method of deep learning target detection to detect fire images, mostly through the convolutional neural network (CNN, Convolutional Neural Networks) to learn the image features of flames, and then when the monitoring image appears in the same When the flame image features learned by the convolutional neural network, it is to detect the presence of flames in the monitored scene. However, in actual scenes, the shape, color, and size of flames in different scenes are very different. Therefore, the existing volume When the product neural network performs flame detection in different scenarios, it is easy to have missed detection and false detection

Images