Smoke detecting method based on random forest characteristic selection

Abstract

The invention discloses a smoke detecting method based on random forest characteristic selection. The method comprises the following steps of (1), determining four features of the smoke in a smoke image as input of a random forest model; (2), synthesizing the smoke image by means of a no-smoke image, and constructing a smoke block sample and a non-smoke block sample by means of an image segmentation method; (3), performing dimension-reducing feature selection through random forest model training for obtaining a regression feature; (4), by means of a support vector machine (SVM) and the regression feature, training a smoke block sample and a non-smoke-block sample, obtaining a classifier; and (5), performing real-time smoke detection, namely performing smoke detection on the image in a video according to the classifier which is obtained by the SVM. The smoke detecting method has advantages of performing in-time smoke early-warning and reducing error warning rate in fire early-warning, obtaining a smoke area without setting an experiment threshold, and automatically performing smoke feature analysis by means of random forest feature selection.

Description

technical field [0001] The invention relates to a smoke detection method based on random forest feature selection, belonging to the technical field of smoke detection. Background technique [0002] Fire accidents will not only cause social and economic losses, but also cause certain damage to the ecological environment, and even seriously threaten people's lives and property safety. Therefore, it is urgent to detect fires to prevent fire accidents. [0003] Traditional fire detection methods are basically based on the single data judgment method of the sensor, and generally use electronic devices or optical devices to detect flames and smoke particles generated during the fire. Although this sensor-based detection method is relatively simple, it has major flaws because of its sensitivity and dependence on the environment. It is easily affected by dust and other particles in the environment, and in order to improve accuracy, it must be close to the fire source. In a large sp...

AI Technical Summary

Problems solved by technology

Smoke detection based on machine learning methods such as BP neural network and SVM combined with extracted smoke features can achieve the purpose of automatic smoke detection and reduce the influence of empirical thresholds, but it is difficult to meet the requirements of automatic analysis of smoke features, and smoke samples The acquisition of non-smoke and non-smoke samples requires a large amount of video data, and it is difficult to extract pure smoke samples

Images