Flame data generation method and terminal

Abstract

The invention discloses a flame data generation method and a terminal. The method comprises the steps of obtaining a training sample set and a preset background image; training a flame generation network by using the training sample set to obtain a trained flame generation network; performing random flame generation on the preset background image by using the trained flame generation network to obtain a preset background image with flame; and inputting the preset background image with the flame into a preset flame segmentation network for flame segmentation to obtain a flame image, and synthesizing the flame image and the preset background image to generate flame data, so that random flame can be generated for the specific preset background image, and an effect that a certain position of an original image is on fire is formed; therefore, the final flame data better conforms to the actual situation and is more real and effective, so the sample data of flame detection is effectively increased, and the problems that the data sample size is insufficient and the flame data lacks diversity are solved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a flame data generation method and a terminal. Background technique [0002] Target detection is one of the most important tasks in computer vision. Among them, the target detection of wildfire has great practical and economic significance, and has important research value. The current difficulty in mountain fire detection lies in: (1) the lack of fire data; (2) the large difference in the characteristics of day and night scenes; (3) the shape of flames is ever-changing, and the characteristics are unstable, which further exacerbates the problem of insufficient data. [0003] At present, most flame detection methods based on deep learning can only train models or improve algorithms on existing data, but the difficulty of this task is not the algorithm and model itself, but the lack of data, and due to the ever-changing flame shape, The number of flames in each form is a...

AI Technical Summary

Problems solved by technology

[0003] At present, most flame detection methods based on deep learning can only train models or improve algorithms on existing data, but the difficulty of this task is not the algorithm and model itself, but the lack of data, and due to the ever-changing flame shape, The number of flames in each form is also uneven; some places have more fires, some places have fewer fires, and in places where there are fewer fires, we cannot get enough training data for the same scene, so in the imperfect dataset Under such circumstances, it is futile to continue to blindly improve algorithms and deep learning models

Images