| CPC G01N 15/1456 (2013.01) [G01N 15/1425 (2013.01); G01N 15/1429 (2013.01); G01N 2015/0046 (2013.01); G01N 2015/1486 (2013.01); G01N 2015/1497 (2013.01)] | 16 Claims |
|
1. A fire smoke detection method based on particle shape characteristics for detecting particles in an optical dark chamber, wherein the optical dark chamber comprises a light source combination and a photoelectric conversion module D, the light source combination is a light source B1, a light source I1, and a light source B2, the light source B1 and the light source B2 are both blue light sources, the light source I1 is an infrared light source, a center of the optical dark chamber is O, an angle of change in an optical path of a light ray from the light source B1 reflected through the center O to the photoelectric conversion module D is an acute angle, an angle of change in an optical path of a light ray from the light source I1 reflected through the center O to the photoelectric conversion module D is an acute angle, and an angle of change in an optical path of a light ray from the light source B2 reflected through the center O to the photoelectric conversion module D is an obtuse angle, wherein scattering characteristics of the light sources B1 and the light sources B2, which have same wavelength but different observation angles, are used to distinguish particle shapes, scattering characteristics of the light source I1, which has a different wavelength, are used to correct the deviation caused by large aerosol particles, extending an aerosol classification curve to a classification surface by combining data from three different light sources, obtaining an accurate classification result by a neural network model, which is used to learn and find coefficients of the classification surface equation,
the fire smoke detection method comprises:
step S1: obtaining a background value of a scattered light optical power of each light source, wherein the background value of the each light source is an optical power of scattered light from corresponding light source received by the photoelectric conversion module in an environment without fire smoke and interfering aerosols;
step S2: periodically starting the light source combination to send light pulse signals in sequence, calculating a change value of the each light source and determining whether any change value exceeds a set threshold value, when not, repeating step S2, and when so, it is considered that there is an abnormality, and executing step S3, wherein the change value of the each light source is an absolute value of a difference value between a current scattered light optical power from the light source received by the photoelectric conversion module D and the background value of the scattered light optical power of the light source;
step S3: starting the light source combination n times in succession after an abnormality occurs and recording a resulting change value, where n is a positive integer; and
step S4: constructing three change values of the light source combination at each start-up into a space vector, classifying n space vectors by a classifier based on differences in particle shape between fire smoke and interfering aerosols, obtaining n classification results, wherein the classifier is a trained neural network model, and classification categories comprise fire smoke and interfering aerosols, the fire smoke exhibits chain-like shape, the interfering aerosols are spherical; counting the obtained n classification results; and when a number of one category is greater than or equal to k, determining that the category is a correct classification result, where k>n/2.
|