Note_Automatic Water-Body Segmentation From High-Resolution Satellite Images via Deep Networks

基本信息

二区文章,水域分割

Automatic Water-Body Segmentation From High-Resolution Satellite Images via Deep Networks

笔记

出发点

  1. 水域分割是遥感的基本任务。
  2. 传统的方法依赖光谱,只能处理分辨率低图像。而分辨率强的图片,包含更多细节。
  3. 不同数据传感器获得数据,方法的鲁棒性得到考验。

主要的创新点

  1. 提出一个新的分割网络RRF DeconvNet 网络(restricted receptive field deconvolution network)
  2. 认为通常loss,也就是L2距离或者说是欧拉距离,不能很好的突出分割中边界的作用,所以,使用了EWLoss,就是高斯加权的欧拉距离。

详细说明

网络结构信息和网络配置。

可以看出,主要的改动是在网络结构上,以及大量使用空洞卷积去代替普通卷积来获得更大的感受野。

使用高斯加权距离作为代价函数,主要的特点想法是突出边际,越靠近分割边界,获得加权越大。

实验

数据是从google earth 上面提取的,0.5M,大小是512*512,主要是四川,武汉地区,总共九千张图片,7:2:1=train:validation:test

主要是六个实验,(两类loss 函数,三个网络)

评价指标也是两个,一个是overrall pixels,一个是Edge pixels。前者是常规的的

\[OP= \frac{TP+TN}{TP+TN+FP+FN}\]

后面一个不是很清晰,大概的描述是这样(具有与边界最大5个棋盘距离的像素被认为是边缘像素):

 the pixels who have a maximum 5 chessboard distance to boundaries are considered as edge pixels

看结果,好像评价函数对结果的影响不大,或者说,基本没什么影响。OP是主流的评价指标,为什么提升不明显,很显然,由于图片的像素太多,边界部分像素点太少,有点样本不平衡的原因,所以就算有所提升,也不会有很明显的变化。所以引入新的评价变量,评价loss的目的是达到了,确实可以让边缘更加准确清晰。

总结

  1. 改变网络结构
  2. 改变loss
  3. 设计有利评价指标
  4. 值得一提,采用的框架是mxnet

原文地址:https://www.cnblogs.com/blog4ljy/p/8955025.html

时间: 2024-11-29 08:15:42

Note_Automatic Water-Body Segmentation From High-Resolution Satellite Images via Deep Networks的相关文章

Review of Semantic Segmentation with Deep Learning

In this post, I review the literature on semantic segmentation. Most research on semantic segmentation use natural/real world image datasets. Although the results are not directly applicable to medical images, I review these papers because research o

【论文笔记】A Review on Deep Learning Techniques Applied to Semantic Segmentation

A Review on Deep Learning Techniques Applied to Semantic Segmentation 2018-02-22  10:38:12   1. Introduction: 语义分割是计算机视觉当中非常重要的一个课题,其广泛的应用于各种类型的数据,如:2D image,video,and even 3D or volumetric data. 最近基于 deep learning 的方法,取得了非常巨大的进展,在语义分割上也是遥遥领先于传统算法. 本

{ICIP2014}{收录论文列表}

This article come from HEREARS-L1: Learning Tuesday 10:30–12:30; Oral Session; Room: Leonard de Vinci 10:30  ARS-L1.1—GROUP STRUCTURED DIRTY DICTIONARY LEARNING FOR CLASSIFICATION Yuanming Suo, Minh Dao, Trac Tran, Johns Hopkins University, USA; Hojj

CVPR 2015 papers

CVPR2015 Papers震撼来袭! CVPR 2015的文章可以下载了,如果链接无法下载,可以在Google上通过搜索paper名字下载(友情提示:可以使用filetype:pdf命令). Going Deeper With ConvolutionsChristian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke

[C6] Andrew Ng - Convolutional Neural Networks

About this Course This course will teach you how to build convolutional neural networks and apply it to image data. Thanks to deep learning, computer vision is working far better than just two years ago, and this is enabling numerous exciting applica

Open Data for Deep Learning

Open Data for Deep Learning Here you'll find an organized list of interesting, high-quality datasets for machine learning research. We welcome your contributions for curating this list! You can find other lists of such datasets on Wikipedia, for exam

生成对抗网络资源 Adversarial Nets Papers

来源:https://github.com/zhangqianhui/AdversarialNetsPapers AdversarialNetsPapers The classical Papers about adversarial nets The First paper ? [Generative Adversarial Nets] [Paper] [Code](the first paper about it) Unclassified ? [Deep Generative Image

阅读笔记 CCL: Cross-modal Correlation Learning with Multi-grained Fusion by Hierarchical Network

总结 CCL: Cross-modal Correlation Learning with Multi-grained Fusion by Hierarchical Network Yuxin Peng, Jinwei Qi, Xin Huang and Yuxin Yuan 常见方法 使用深度神经网络(DNN)的跨模态检索大体分为两个步骤: 1 The first learning stage is to generate separate representation for each mo

CVPR 2017 Paper list

CVPR2017 paper list Machine Learning 1 Spotlight 1-1A Exclusivity-Consistency Regularized Multi-View Subspace Clustering Xiaojie Guo, Xiaobo Wang, Zhen Lei, Changqing Zhang, Stan Z. Li Borrowing Treasures From the Wealthy: Deep Transfer Learning Thro