Timeline

1. 论文写作背景

农田具有小块、破碎、形状不规则的特点，且免费卫星图像分辨率不足，导致边界识别困难； 现有深度学习模型需人工逐图标注农田（Labeling），耗时费力； 热门的“分割一切模型”（Segment Anything Model，SAM）虽强，但仅适用于高分辨率的简单图像，对卫星图适配性差。

2. 核心方法：SIDEST 框架

作者提出的 SIDEST 框架分为三步：

（1）超分辨率重建（Super-Resolution，SR）—— 给卫星图“戴眼镜”

• 技术原理：采用生成对抗网络（GAN），先将高清图人工模糊为“低清图”，训练 AI 学习从“低清图”还原“高清图”的能力。
• 效果：将 10 米分辨率的卫星图，通过算法“脑补”成 2.5 米的高清图，使农田边界更清晰。

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.

A sticky article demo.

Heading 2

Here is the content.

Heading 3

Here is the content.

Heading 2

Here is the content.

Heading 3

Here is the content.