Remote sensing image smoke detection method based on feature fusion

Abstract

The invention discloses a feature fusion-based remote sensing image smoke detection method. A feature fusion-based convolutional neural network provided by the invention mainly comprises three parts: a backbone network, a global feature branch and a local feature branch, wherein the backbone network is used as a preliminary feature extractor; the global feature branch is used for extracting global features such as repeated textures; the local branches are used for extracting significant features in the local key area; smoke in the remote sensing image has different shapes, textures and ranges, and smoke detection is facilitated by considering global features and local features. According to the remote sensing smoke detection method disclosed by the invention, the highest detection precision 96.22% is obtained on a public remote sensing data set.

Description

technical field [0001] The invention belongs to the field of video image processing, and in particular relates to a method for detecting smoke in remote sensing images based on feature fusion. Background technique [0002] Fire is a common natural disaster, which will cause great harm to human safety and property. It is very important to fire detection and alarm. When a fire occurs, the smoke captured by the satellite is used as a signal of the fire. With the increasing availability and performance of satellite remote sensing technology, the use of remote sensing images to detect smoke for fire detection has been widely adopted. However, the shape, texture, and color of smoke in remote sensing images vary greatly. Moreover, some scenes in remote sensing data, such as clouds and haze, are very similar to smoke, which increases the difficulty of smoke detection. Therefore, the research on remote sensing smoke detection is of great significance. [0003] With the developmen...

AI Technical Summary

Problems solved by technology

Many existing convolutional neural networks focus on global features based on repeated textures, while ignoring local features in images

Images