All Projects → anxiangsir → Deeplabv3 Tensorflow

anxiangsir / Deeplabv3 Tensorflow

使用deeplab_v3模型对遥感图像进行分割

Programming Languages

python
139335 projects - #7 most used programming language
python3
1442 projects

Projects that are alternatives of or similar to Deeplabv3 Tensorflow

massive-change-detection
QGIS 2 plugin for applying change detection algorithms on high resolution satellite imagery
Stars: ✭ 18 (-93.33%)
Mutual labels:  remote-sensing
neural-road-inspector
After a hurricane, roads are often flooded or washed out, making them treacherous for motorists. Using state of the art deep learning methods, I attempted to automatically annotate flooded, washed out, or otherwise severely damaged roads. My goal is create a tool that can help detect and visualize anomalous roads in a simple user interface.
Stars: ✭ 37 (-86.3%)
Mutual labels:  remote-sensing
lsru
🔍 🌐Query and Order Landsat Surface Reflectance data via ESPA
Stars: ✭ 24 (-91.11%)
Mutual labels:  remote-sensing
biodivMapR
biodivMapR: an R package for α- and β-diversity mapping using remotely-sensed images
Stars: ✭ 18 (-93.33%)
Mutual labels:  remote-sensing
FDCNN
The implementation of FDCNN in paper - A Feature Difference Convolutional Neural Network-Based Change Detection Method
Stars: ✭ 54 (-80%)
Mutual labels:  remote-sensing
NodeMICMAC
A Lightweight REST API to Access MICMAC Photogrammetry and SFM Engine.
Stars: ✭ 54 (-80%)
Mutual labels:  remote-sensing
land-cover-to-land-use-classification
Satellite image processing pipeline to classify land-cover and land-use
Stars: ✭ 64 (-76.3%)
Mutual labels:  remote-sensing
Torchsat
🔥TorchSat 🌏 is an open-source deep learning framework for satellite imagery analysis based on PyTorch.
Stars: ✭ 261 (-3.33%)
Mutual labels:  remote-sensing
open-impact
To help quickstart impact work with Satellogic [hyperspectral] data
Stars: ✭ 21 (-92.22%)
Mutual labels:  remote-sensing
shipsnet-detector
Detect container ships in Planet imagery using machine learning
Stars: ✭ 30 (-88.89%)
Mutual labels:  remote-sensing
moveVis
An R package providing tools to visualize movement data (e.g. from GPS tracking) and temporal changes of environmental data (e.g. from remote sensing) by creating video animations.
Stars: ✭ 104 (-61.48%)
Mutual labels:  remote-sensing
Landsat578
Very simple API to download Landsat [1-5, 7, 8] data from Google
Stars: ✭ 54 (-80%)
Mutual labels:  remote-sensing
ms-convSTAR
[RSE21] Pytorch code for hierarchical time series classification with multi-stage convolutional RNN
Stars: ✭ 17 (-93.7%)
Mutual labels:  remote-sensing
modape
MODIS Assimilation and Processing Engine
Stars: ✭ 19 (-92.96%)
Mutual labels:  remote-sensing
geowombat
GeoWombat: Utilities for geospatial data
Stars: ✭ 34 (-87.41%)
Mutual labels:  remote-sensing
GGHL
This is the implementation of GGHL (A General Gaussian Heatmap Label Assignment for Arbitrary-Oriented Object Detection)
Stars: ✭ 309 (+14.44%)
Mutual labels:  remote-sensing
RAMS
Official TensorFlow code for paper "Multi-Image Super Resolution of Remotely Sensed Images Using Residual Attention Deep Neural Networks".
Stars: ✭ 55 (-79.63%)
Mutual labels:  remote-sensing
Otb
Github mirror of https://gitlab.orfeo-toolbox.org/orfeotoolbox/otb
Stars: ✭ 265 (-1.85%)
Mutual labels:  remote-sensing
ecmwf models
Python package for downloading ECMWF reanalysis data and converting it into a time series format.
Stars: ✭ 27 (-90%)
Mutual labels:  remote-sensing
geoblaze
Blazing Fast JavaScript Raster Processing Engine
Stars: ✭ 80 (-70.37%)
Mutual labels:  remote-sensing

基于Deeplab-v3对遥感图像的语义分割

QQ:1363090802

数据集:

CCF卫星影像的AI分类与识别提供的数据集初赛复赛训练集,一共五张卫星遥感影像

dataset
├── origin //5张遥感图片,有标签
├── test   //3张遥感图片,无标签,在这个任务中没有用到
└── train  //为空,通过`python preprocess.py`随机采样生成
    ├── images       
    └── labels

其中我们使用前四张用来做训练,最后一张用来做测试

dependency

cuda==8.0
cudnn==6
python==3.5
pip install opencv-python==3.4.2.17
pip install tensorflow-gpu==1.13.1
pip install sklearn pip install pandas

主要策略:

  • [x] 将原始的遥感图像裁成大小为(256x256)的图片块,裁剪的方法为随机采样,并进行数据扩增
  • [x] 搭建Deeplab-v3模型,使用预训练的 resnet-v2-50 迁移学习
  • [x] 完整的训练测试程序,使用 tensorboard 监控模型训练
  • [x] 多尺度拼接预测,提升模型
  • [ ] 后处理优化,比如消除预测图片拼接痕迹
  • [ ] 使用更好的骨干网络,如 Xception

最终结果:

评价方法为 mean-IoU,在数据集极少的情况下,测试集评价结果得到了 77.3 的分数

方法 mean-IoU accuracy
baseline(deeplabv3) 71.2 -
resnet-v2-50 pretrain 77.1 -
旋转四次预测取平均 77.6 85.5

如何训练

将百度云中的数据集文件夹dataset下载并存放到项目主目录下
python proprecess.py 时间稍长,需要等待
python train.py 时间稍长,可以更改args.test_display 多久查看一次测试结果

如何可视化训练过程

cd 到主目录下
tensorboard --logdir=./

测试结果:

模型预测step = 10000 模型预测step = 50000
测试图片缩略图 测试图片标签上色图
tensorboard
Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].