Smoke detection is very important in early-warning of forest-fire. Because of the limitations of lower anti-interference ability

traditional technology is hard to distinguish fire smoke

water fog

the grass and other false information. Therefore

a new smoke detection method based on multispectral separation is proposed. First

band pass filters in the region of 460~520 nm

540~570 nm

and 580~610 nm were used to seize the principal components in spectra emitted by the things under monitoring. Then

plane of the transform domain was build from these principal components

with simple algebraic judgment on the plane

classification and recognition of the smoke

fog and grass were performed. Finally

with as many as 1000 times Monte Carlo simulation

the three-band spectral analysis method proposed has the maximum separability

and easy operation. The proposed method could overcome the shortcoming which cannot distinguish smoke and water fog in traditional image and video detection methods

it no longer needs complex spectra classification algorithms

has certain practical value in the field of fire monitoring.