Refinery fire identification method based on convolutional neural network

Abstract

The invention relates to a refinery fire identification method based on a convolutional neural network, and belongs to the technical field of fire video identification. The method comprises the following steps of: 1, detecting a running target through a background difference method, and judging whether a moving object exists in a current picture or not; 2, transmitting the moving target into a flame detection model, and outputting a flame probability and a flame region; 3, transmitting the suspicious area into a false alarm model, and outputting the probability of real flames; and 4, judging whether flame or flame false alarm exists according to the probability of the real flame. Through verification, the method has higher recognition accuracy and lower false alarm rate, and the model can be updated according to the actually running picture, so that the method is more suitable for the actual scene of the scene; through real-time monitoring of a control area, a rapid response can be made at the initial stage of a fire, and real-time analysis and post-processing are carried out, so that early forecasting and control of the fire can be realized, and the influence and loss caused by the fire can be reduced.

Description

technical field [0001] The invention relates to a fire recognition method in a refinery based on a convolutional neural network, and belongs to the technical field of fire video recognition. Background technique [0002] The traditional image detection technology of flame features adopts the method of manually designing features to extract the static and dynamic features of the flame through the specific description of the design, and then uses these obvious flame features for flame recognition. This technical solution cannot learn the inherent advanced features of the flame, the recognition accuracy is low, and the false alarm rate is high, and it will misreport a variety of suspected flame objects as fire. Frequent flame alarms in actual use will cause too much interference to users, and users will gradually reduce their attention to alarms. When a flame actually occurs, it will not be dealt with in time. The figure below is the feature construction process of the traditi...

AI Technical Summary

Problems solved by technology

This technical solution cannot learn the inherent advanced features of the flame, the recognition accuracy is low, the false alarm rate is high, and various suspected flame objects will be misreported as fire

Frequent flame alarms in actual use will cause too much interference to users, and users will gradually reduce their attention to alarms. When a flame actually occurs, it will not be dealt with in time

Images