基于频率Transformer-CNN耦合的烟雾分割模型研究

Research on Smoke Segmentation Model Based on Frequency Transformer CNN Coupling

本文深入开展了输电线路附近山火实时监测过程中图像的烟雾分割方法研究,有助于对图像中烟雾体积、扩散方向和源头等准确提取信息,这对制定应急预案具有重要意义。为此,提出了一种名为CFTNet的双分支分割模型。该模型将频率Transformer分支与CNN分支结合起来,优化了全局和局部特征的表示。此外,本文还设计了一个混合自注意力融合模块(HSAM),以高效地融合来自频率Transformer分支和CNN分支的信息。研究表明,该算法的性能优于其他主流分割方法。

This article delves into the study of smoke segmentation methods in real-time monitoring of wildfires near transmission lines,which helps to accurately extract information such as smoke volume,diffusion direction,and source from images.This is of great significance for developing emergency plans.To this end,a dual branch segmentation model called CFT-Net was proposed in the study.This model combines the frequency Transformer branch with the CNN branch to optimize the representation of global and local features.In addition,this article also designs a hybrid self attention fusion module(HSAM)to efficiently fuse information from the frequency Transformer branch and the CNN branch.Research has shown that the performance of this algorithm is superior to other mainstream segmentation methods.