[论文笔记]Remote Sensing Image Change Detection with Transformers

Remote Sensing Image Change Detection with Transformers

论文简介:

该论文是2021年发表在ISPRS上的利用transformer做变化检测的工作。利用了LEVIR-CD数据集,下图是论文提出的BIT在该数据集上的结果。请添加图片描述

网络模型:

在这里插入图片描述

关键点:

BIT能够学习并且关联全局高层语义的上下文信息并可以反馈附加到原始特征。

3个主要步骤

  1. 从CNN backbone中提取高层特征。
  2. 用BIT来修饰双时图像特征
  3. 由预测头获得变化图。

算法流程图在这里插入图片描述

3个关键组件

1.a Siamese semantic tokenizer, 2. transformer encoder (TE),3. a Siamese TD。

a Siamese semantic tokenizer:

在这里插入图片描述
在这里插入图片描述

TE:

这一模块,在已经获得了两个语义token集。
该模块核心思想是MSA,多个独立注意力分支并行计算。MSA 的优点是它可以在不同的位置联合处理来自不同表示子空间的信息。
在这里插入图片描述

TD:

经过TE后,得到了两个语义丰富tokens集,这些语义丰富tokens包含紧凑的高级语义信息,很好地揭示了区域变化信息。 在TD结构中我们需要投影回到像素空间以获得像素级特征。论文认为每个像素都可以用紧致语义的tokens组合来表示。
在这里插入图片描述

网络细节:

1)CNN主干网络:修改的ResNet18来提取双时图像特征图。ResNet18一共有五个阶段,每个下采样倍率为2。修改最后两个阶段倍率为1,在最后添加卷积层减少输出通道至32。为了减少空间细节的损失,后加双线性插值层获得下采样倍率为4。
2)双时Transformer: token length L = 4 、NTE=1、NTD=8、h=8、d=8。
3)在这里插入图片描述
4)
在这里插入图片描述

  • 1
    点赞
  • 5
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
Pest detection based on UAV remote sensing images is an emerging field that has gained significant attention in recent years. UAVs provide a unique advantage in pest detection as they can cover large areas quickly and efficiently, and collect high-resolution images that enable detailed analysis of crop health. The process of pest detection using UAV remote sensing images involves several steps. First, the UAV is equipped with a multispectral camera that can capture images in different wavelengths. These images are then processed using computer vision algorithms to identify areas of the crop that may be affected by pests or diseases. The algorithms can analyze the images to identify changes in the color, texture, and shape of the plants, which may indicate the presence of pests or diseases. The data collected by the UAV is then used to generate maps that show the extent and severity of the pest infestation. The use of UAVs in pest detection has several advantages over traditional methods. UAVs can cover large areas quickly and efficiently, which saves time and resources. They can also collect high-resolution images that enable detailed analysis of crop health, which can help farmers make more informed decisions about pest management. Overall, pest detection based on UAV remote sensing images has the potential to revolutionize the way we monitor and manage pest infestations in crops. As technology continues to advance, we can expect to see more sophisticated algorithms and sensors that can detect pests and diseases with even greater accuracy and efficiency.
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值