Estimation model training method, estimation method and system of forest fire source

Abstract

The invention discloses a forest fire source estimation model training method, estimation method and system, and belongs to the fields of computer vision and forest pyrotechnic video monitoring. The estimation model of the present invention includes a backbone network, a foreground feature enhancement module, and a multi-frame feature fusion module. When training the model, it includes constructing a forest fire video sample data set; preprocessing positive samples and negative samples; Input the video sample into the backbone network to extract the high-dimensional features of the image, input the high-dimensional features into the foreground feature enhancement module to enhance the foreground features, input the multi-frame feature fusion module for feature fusion, perform deconvolution after fusion, and map the high-dimensional features into fire Estimation mask at the source; model parameter optimization using the Focal Loss loss function. The estimation method uses the trained model to estimate the forest fire source, and deploys the estimation method in the estimation system. The invention can accurately identify and estimate the forest fire source.

Description

technical field [0001] The invention belongs to the field of computer vision and forest pyrotechnic video monitoring, and in particular relates to an estimation model training method, estimation method and system for the source of forest fires. Background technique [0002] Forest fire refers to the behavior of forest fire that loses human control, spreads in a large range of forest areas, and causes a lot of losses to forest ecosystems. Forest fires are sudden and destructive, and are one of the most frequent, most destructive, and most difficult natural disasters in the world. They are also the biggest threat to forest resource security. my country is generally an ecologically fragile country and a country with frequent forest fires. The task of forest fire prevention is the primary task of national forest resource protection, which is related to the safety of people's lives and property and forest resources. In recent years, with the improvement of computer performance a...

AI Technical Summary

Problems solved by technology

[0005] Technical problem: The present invention aims at the problem that the existing forest fire identification method is difficult to accurately identify and estimate the fire source, and provides an estimation model training method for the forest fire source, and uses the estimation model trained by the training method to provide A method and system for estimating the source of forest fires. The invention can estimate the source of forest fires, thereby accurately determining the location of the fire and timely extinguishing the fire.

Images