|
| 1 | +<!DOCTYPE html> |
| 2 | +<html lang="en"> |
| 3 | + <head> |
| 4 | + <meta charset="UTF-8" /> |
| 5 | + <meta name="viewport" content="width=device-width, initial-scale=1.0" /> |
| 6 | + <title>UN Open GIS</title> |
| 7 | + <meta name="author" content="UN Open GIS" /> |
| 8 | + <meta |
| 9 | + name="description" |
| 10 | + content="The official site for the UN Open GIS Initiative." |
| 11 | + /> |
| 12 | + <!-- reset default CSS --> |
| 13 | + <link |
| 14 | + rel="stylesheet" |
| 15 | + href="https://cdnjs.cloudflare.com/ajax/libs/normalize/8.0.1/normalize.min.css" |
| 16 | + /> |
| 17 | + <link |
| 18 | + rel="stylesheet" |
| 19 | + href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css" |
| 20 | + /> |
| 21 | + <link rel="stylesheet" href="style.css" /> |
| 22 | + <script defer src="script.js"></script> |
| 23 | + </head> |
| 24 | + |
| 25 | + <body> |
| 26 | + <div id="header"></div> |
| 27 | + |
| 28 | + <section class="wrapper"> |
| 29 | + <article class="content-main"> |
| 30 | + <article class="sub-nav"> |
| 31 | + <a href="index.html" |
| 32 | + ><img src="./Overview_files/sub-nav-home.png" alt="home-icon" |
| 33 | + /></a> |
| 34 | + <u |
| 35 | + ><img src="./Overview_files/sub-nav-arrow.png" alt="arrow-icon" |
| 36 | + /></u> |
| 37 | + <span>Best Practices</span> |
| 38 | + </article> |
| 39 | + <article class="sub-title"> |
| 40 | + <p>Best Practices</p> |
| 41 | + </article> |
| 42 | + <div class="gray-bar"></div> |
| 43 | + <table class="no-border"> |
| 44 | + <tr> |
| 45 | + <th>Subject</th> |
| 46 | + <td> |
| 47 | + [Best Practice] Mapping Refugee Settlements by using |
| 48 | + PulseSatellite |
| 49 | + </td> |
| 50 | + </tr> |
| 51 | + <tr> |
| 52 | + <th>Name</th> |
| 53 | + <td>Secretariat</td> |
| 54 | + </tr> |
| 55 | + <tr> |
| 56 | + <th>Date</th> |
| 57 | + <td>2020-02-12</td> |
| 58 | + </tr> |
| 59 | + <tr> |
| 60 | + <th>Views</th> |
| 61 | + <td>2,722</td> |
| 62 | + </tr> |
| 63 | + </table> |
| 64 | + <div class="gray-bar"></div> |
| 65 | + |
| 66 | + <section class="underline"> |
| 67 | + <div class="bp1-content"> |
| 68 | + <p class="note"> |
| 69 | + <img src="./BP_1_files/202002121352284824.jpg" alt="BP_1: pic" /> |
| 70 | + <span class="caption" |
| 71 | + >Figure 1. Refugee Camp mapping using PulseSatellite</span |
| 72 | + > |
| 73 | + </p> |
| 74 | + <p> |
| 75 | + Counting and classifying structures in a refugee settlement is a |
| 76 | + common analysis task for humanitarian agencies. In practice, this |
| 77 | + is currently manually done by human expert analysts using |
| 78 | + satellite imagery. A single settlement may have tens of thousands |
| 79 | + of structures, and identifying each of them can take several days. |
| 80 | + For camp mapping, PulseSatellite uses a Mask R-CNN model (He et |
| 81 | + al. 2017) trained on images from 12 settlements, where the image |
| 82 | + of each has been split into 300x300 pixel tiles and annotated by |
| 83 | + human experts. Once the model has been run on an unseen camp, the |
| 84 | + analyst can inspect the result in the tool and correct the outputs |
| 85 | + on a subset of tiles. An adaptation stage can then be performed to |
| 86 | + fine-tune the model to the unseen image and increase performance. |
| 87 | + In recent independent tests camp completion rates increased from |
| 88 | + 77.3% to 94.7% after adaptation, with a final user accuracy of |
| 89 | + 94.4% -in line with humanitarian performance requirements |
| 90 | + described in (Quinn et al. 2018). |
| 91 | + </p> |
| 92 | + </div> |
| 93 | + |
| 94 | + <div class="bp1-content"> |
| 95 | + <p class="note"> |
| 96 | + <img |
| 97 | + src="./BP_1_files/Uploadify_202002121349058190.jpg" |
| 98 | + alt="BP_1_2: pic " |
| 99 | + /> |
| 100 | + <span class="caption"> |
| 101 | + Figure2. PulseSatellite user-interface showing a hierarchy of |
| 102 | + learning models |
| 103 | + </span> |
| 104 | + </p> |
| 105 | + <p> |
| 106 | + PulseSatellite is a distributed cloud application, currently |
| 107 | + running on Amazon Web Services (AWS). The web-based user interface |
| 108 | + (UI) is written in JavaScript+HTML, and the web server, which |
| 109 | + communicates with the UI via HTTP and WS protocols, is written in |
| 110 | + NodeJS. The database is managed using PostgreSQL with PostGIS |
| 111 | + extension and is currently also used as a message streaming |
| 112 | + service for Pub/Sub messaging between the web, GPU and tiling |
| 113 | + servers. All backend files are generated and stored on a Amazon |
| 114 | + Elastic File System (EFS). Multiple GPU servers can run ML |
| 115 | + processes across a range of frameworks including PyTorch, |
| 116 | + TensorFlow and Keras. GIS processes, including the tiling servers, |
| 117 | + use OSGeo libraries and programs, most notably GDAL. All used |
| 118 | + software is open source. |
| 119 | + </p> |
| 120 | + <p class="note"> |
| 121 | + * Note: The WG5 Best Practice "Mapping Refugee Settlements by |
| 122 | + using PulseSatellite" has extracted from a publication of |
| 123 | + <a href="https://arxiv.org/pdf/2001.10685" target="_blank" |
| 124 | + >Tomaz Logar et al. 2020</a |
| 125 | + >. |
| 126 | + </p> |
| 127 | + </div> |
| 128 | + </section> |
| 129 | + </article> |
| 130 | + </section> |
| 131 | + <div class="bar"></div> |
| 132 | + |
| 133 | + <div id="footer"></div> |
| 134 | + </body> |
| 135 | +</html> |
0 commit comments