Projects - User contributions [en]

Problem Space

2018-08-30T15:04:38Z

A1669892: Created page with "Needs to be copied from confluence..."

Needs to be copied from confluence...

HR Image Registration

2018-08-30T14:58:37Z

A1669892: Created page with " <p>Image registration has been performed using the open source tool-set QGIS with its add on Georeference GDAL. The process to obtain the final aligned images are as..."

<p>Image registration has been performed using the open source tool-set QGIS with its add on Georeference GDAL. The process to obtain the final aligned images are as follows:</p><ol><li>In QGIS, obtain intersection of both HR images by using the <em>Clipper </em>tool found under <em>Raster → </em><em>Extraction</em> on the top toolbar. This here allows for a bounding box to be selected of the desired area.</li><li>Using the <em>Clipper </em>tool again, divide the images up into 4 tiles. These tiles will be used with the <em>Georeferencer </em>tool. The need for dividing into sub-images is due to the Ground Control Points (GCP) needing to be plentiful in each georeferenced area. The alignment works by aligning these points of the moving and reference image and performing a transformation on the final image hence a smaller image results in more possible GCP's.</li><li>Open <em>Georeferencer </em>tool under <em>Raster → Georeferencer</em> in the top toolbar. For each of the 4 sub-images of the aligning HR image, perform the following tasks:<ol><li>Load in one of the 4 aligning sub-image.</li><li>With <em>add point </em>selected, select a point (usually of a structure) which can be used as a GCP. After choosing a point, select in the appearing window <em>From map canvas.</em> From here, select with the best approximation possible the same point on the reference image. Then, click okay and confirm this selection.</li><li>Repeat step b. ~10 times for approximately evenly spaced points on the image. </li><li>Now, select <em>Settings → Transformation settings</em>. Here, select <em>Helmert</em> transformation, <em>Cubic </em>resampling method and choose output raster directory. This will save the transformed aligned raster.<br/><strong>*<em>Note: Other transformations available including Poly, Projective etc. were all attempted. Helmert provided best result.</em></strong></li></ol></li><li>Repeat step 3 4 times for the 4 tiles. After this is completed, go to <em>Raster → Miscellaneous → Merge</em> and merge these sub-images into 1 full image to obtain the final, aligned HR image.</li></ol><p>The used GCP point data that can be loaded into the <em>Georeferencer</em> are found here: <a href="/confluence/download/attachments/75139161/top_1.tiff.points?version=1&modificationDate=1529755767976&api=v2" data-linked-resource-id="78711113" data-linked-resource-version="1" data-linked-resource-type="attachment" data-linked-resource-default-alias="top_1.tiff.points" data-linked-resource-content-type="application/octet-stream" data-linked-resource-container-id="75139161" data-linked-resource-container-version="5">top-subimage-1.points, </a><a href="/confluence/download/attachments/75139161/top_2.tiff.points?version=1&modificationDate=1529755768095&api=v2" data-linked-resource-id="78711114" data-linked-resource-version="1" data-linked-resource-type="attachment" data-linked-resource-default-alias="top_2.tiff.points" data-linked-resource-content-type="application/octet-stream" data-linked-resource-container-id="75139161" data-linked-resource-container-version="5">top-subimage-2.points</a>, <a href="/confluence/download/attachments/75139161/bottom_3.tiff.points?version=1&modificationDate=1529755767752&api=v2" data-linked-resource-id="78711111" data-linked-resource-version="1" data-linked-resource-type="attachment" data-linked-resource-default-alias="bottom_3.tiff.points" data-linked-resource-content-type="application/octet-stream" data-linked-resource-container-id="75139161" data-linked-resource-container-version="5">bottom-subimage.points</a>, <a href="/confluence/download/attachments/75139161/bottom_4.tiff.points?version=1&modificationDate=1529755767869&api=v2" data-linked-resource-id="78711112" data-linked-resource-version="1" data-linked-resource-type="attachment" data-linked-resource-default-alias="bottom_4.tiff.points" data-linked-resource-content-type="application/octet-stream" data-linked-resource-container-id="75139161" data-linked-resource-container-version="5">bottom-subimage-3.points</a></p><p><strong>Experimentation with results:</strong></p><p>In pursuit of more consistent alignment of structures within collected image data, experimentation of using 10 more GCP points per sub-image was conducted (20 GCP total points/sub-image). This resulted in the following interesting observations:</p><ul><li>Some areas containing structures suffered a worsened registration compared to the previous 10 GCP point alignment. Although, improvement is seen in the majority of the image.</li><li>Areas containing vineyards and other structures with simple, consistent geometric structures showed to contain almost <strong>No</strong><em> </em><strong>error</strong> after alignment. This was not an observed detail previously noted. When zooming in and performing visual inspection of vineyards across the ROI, they contain no unwanted translation from visual inspection. This is a very good result.</li></ul>

Docker Image

2018-08-30T14:57:35Z

A1669892: Created page with "<p>Docker is a container platform for software production and development. Using containers allows us to ensure a consistent environment of dependencies despite possible discr..."

<p>Docker is a container platform for software production and development. Using containers allows us to ensure a consistent environment of dependencies despite possible discrepancies in running hardware. Docker is well supported across desktop and cloud, which also enables us to develop locally initially and easily upgrade to more powerful cloud based hardware in the future. As our docker container only contains a selection of publicly available software libraries, we have opted to store a compiled publicly on hub.docker.com.</p><p>Our compiled docker image can be pulled using the command:</p><div class="confluence-information-macro confluence-information-macro-information conf-macro output-block" data-hasbody="true" data-macro-name="info"><span class="aui-icon aui-icon-small aui-iconfont-info confluence-information-macro-icon"> </span><div class="confluence-information-macro-body"><p>docker pull jj11teen/project-109</p></div></div><p>The current dockerfile to build the image yourself can be found here: <a href="/confluence/download/attachments/75138166/Dockerfile?version=2&modificationDate=1525240950950&api=v2" data-linked-resource-id="75138179" data-linked-resource-version="2" data-linked-resource-type="attachment" data-linked-resource-default-alias="Dockerfile" data-linked-resource-content-type="application/x-upload-data" data-linked-resource-container-id="75138166" data-linked-resource-container-version="3">Dockerfile</a></p><p>The current packages in the image are:</p><ul><li>tensorflow</li><li>keras</li><li>gdal</li></ul>

Data Wrangling (Low Res)

2018-08-30T14:56:59Z

A1669892:

<h2 id="DataWrangling(LowRes)-LowResolutionAreacoverage:">Low Resolution Area coverage:</h2><p>Res: 20m/pixel, 2866x1651</p><p><span class="confluence-embedded-file-wrapper confluence-embedded-manual-size"></span></p><p>Image shows corners of HR overlapping area we are using (small overlaid images) on LR image taken from Sentinel2 satellite (zoom=12). Pre-processing will be performed to crop unwanted LR area and tile in ROI of HR overlapped images.</p><p>LR image bands available: B1-12, B8A and RGB (will investigate performance effect of using B1-12 vs RGB)</p><h2 id="DataWrangling(LowRes)-LowResolutionImages">Low Resolution Images</h2><div class="table-wrap"><table class="wrapped confluenceTable"><colgroup><col style="width: 139.0px;"/><col style="width: 239.0px;"/></colgroup><tbody><tr><th class="confluenceTh">Date vs Image</th><th class="confluenceTh"><span>lat=-33.8570&lng=115.1865</span></th></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2017-04-27" class="date-past">27 Apr 2017</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/drive/folders/1IUDy6EKRfEz1EoahM0X5XnRviakGOJhs?usp=sharing" class="external-link" rel="nofollow">LR_27-04-2017</a></td></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2017-02-26" class="date-past">26 Feb 2017</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/drive/folders/1wVeK6PqX5uMYZvUYwpDFIlgTt2DruMxN?usp=sharing" class="external-link" rel="nofollow">LR_26-02-2017</a></td></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2017-02-16" class="date-past">16 Feb 2017</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/drive/folders/1ER7Z5wwnOdv0VYoX7Tr6r_6JHL0WZtlr?usp=sharing" class="external-link" rel="nofollow">LR_12-02-2017</a></td></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2016-12-08" class="date-past">08 Dec 2016</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/open?id=1vMb-3EPGUFw6B9CzKgIh1XG9CxrG5mV2" class="external-link" rel="nofollow">LR_08-12-2016</a></td></tr></tbody></table></div><p>The HR images we are using were are taken on <time datetime="2017-01-15" class="date-past">15 Jan 2017</time> and <time datetime="2017-03-31" class="date-past">31 Mar 2017</time></p><p><br/></p>

Data Wrangling (Low Res)

2018-08-30T14:55:41Z

A1669892: Created page with "<h2 id="DataWrangling(LowRes)-LowResolutionAreacoverage:">Low Resolution Area coverage:</h2><p>Res: 20m/pixel, 2866x1651</p><p><span class="confluence-embedded-file-wrapper co..."

<h2 id="DataWrangling(LowRes)-LowResolutionAreacoverage:">Low Resolution Area coverage:</h2><p>Res: 20m/pixel, 2866x1651</p><p><span class="confluence-embedded-file-wrapper confluence-embedded-manual-size"></span></p><p>Image shows corners of HR overlapping area we are using (small overlaid images) on LR image taken from Sentinel2 satellite (zoom=12). Pre-processing will be performed to crop unwanted LR area and tile in ROI of HR overlapped images.</p><p>LR image bands available: B1-12, B8A and RGB (will investigate performance effect of using B1-12 vs RGB)</p><h2 id="DataWrangling(LowRes)-LowResolutionImages">Low Resolution Images</h2><div class="table-wrap"><table class="wrapped confluenceTable"><col style="width: 239.0px;"/></colgroup><tbody><tr><th class="confluenceTh">Date vs Image</th><th class="confluenceTh"><span>lat=-33.8570&lng=115.1865</span></th></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2017-04-27" class="date-past">27 Apr 2017</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/drive/folders/1IUDy6EKRfEz1EoahM0X5XnRviakGOJhs?usp=sharing" class="external-link" rel="nofollow">LR_27-04-2017</a></td></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2017-02-26" class="date-past">26 Feb 2017</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/drive/folders/1wVeK6PqX5uMYZvUYwpDFIlgTt2DruMxN?usp=sharing" class="external-link" rel="nofollow">LR_26-02-2017</a></td></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2017-02-16" class="date-past">16 Feb 2017</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/drive/folders/1ER7Z5wwnOdv0VYoX7Tr6r_6JHL0WZtlr?usp=sharing" class="external-link" rel="nofollow">LR_12-02-2017</a></td></tr><tr><th colspan="1" class="confluenceTh"><div class="content-wrapper"><p><time datetime="2016-12-08" class="date-past">08 Dec 2016</time> </p></div></th><td colspan="1" class="confluenceTd"><a href="https://drive.google.com/open?id=1vMb-3EPGUFw6B9CzKgIh1XG9CxrG5mV2" class="external-link" rel="nofollow">LR_08-12-2016</a></td></tr></tbody></table></div><p>The HR images we are using were are taken on <time datetime="2017-01-15" class="date-past">15 Jan 2017</time> and <time datetime="2017-03-31" class="date-past">31 Mar 2017</time></p><p><br/></p>

Data Wrangling (High Res)

2018-08-30T14:54:21Z

A1669892: Created page with "<p>Here is a Powershell script to download the tiled images from the Consilium VM: The private key needs to be named p2kaggle.pem in the local directory. Input parameters are..."

<p>Here is a Powershell script to download the tiled images from the Consilium VM: The private key needs to be named p2kaggle.pem in the local directory. Input parameters are:</p><ul><li>image: '<span style="color: rgb(255,0,255);">104001002B124500</span>' or '<span style="color: rgb(0,255,255);">10400100265ED400</span>'</li><li>tile: the tile number ranging from 0 to 49 for <span style="color: rgb(23,43,77);text-decoration: none;">'</span><span style="color: rgb(23,43,77);text-decoration: none;">10400100265ED400'</span> and 0 to 160 <span style="color: rgb(23,43,77);text-decoration: none;">'104001002B124500'</span></li></ul><div class="code panel pdl conf-macro output-block" style="border-width: 1px;" data-hasbody="true" data-macro-name="code"><div class="codeContent panelContent pdl">
<pre class="syntaxhighlighter-pre" data-syntaxhighlighter-params="brush: powershell; gutter: false; theme: Confluence" data-theme="Confluence">.\download_images.ps1 -image 104001002B124500 -tile 64</pre>
</div></div><p><span style="color: rgb(23,43,77);text-decoration: none;"><br/></span></p><p>From trial and error the tiles of overlap are:</p><ul><li><span style="color: rgb(23,43,77);text-decoration: none;"><span style="color: rgb(255,0,255);">104001002B124500<span style="color: rgb(23,43,77);"> - </span>Magenta</span> - <span style="color: rgb(0,51,102);text-decoration: none;">2017-03-31T03:00:47.616Z</span></span><ul><li><span style="color: rgb(23,43,77);text-decoration: none;">80, 81, 82, 83; 88, 89, 90, 91; 96, 97, 98, 99; 104, 105, 106, 107; 112, 113, 114, 115; 120, 121, 122, 123; 128, 129, 130, 131;</span></li></ul></li><li><span style="color: rgb(23,43,77);text-decoration: none;"><span style="color: rgb(0,255,255);">10400100265ED400<span style="color: rgb(23,43,77);"> -</span> Cyan</span> - <span style="color: rgb(0,51,102);text-decoration: none;">2017-01-15T02:50:50.846Z</span></span><ul><li>3, 4, 5, 6, 7; 11, 12, 13, 14, 15; 19, 20, 21, 22, 23; 27, 28, 29, 30, 31; 35, 36, 37, 38, 39; 43, 44, 45, 46, 47; 51, 52, 53, 54, 55; 59, 60, 61, 62, 63;<span style="color: rgb(23,43,77);text-decoration: none;"><br/></span></li></ul></li></ul><p><span style="color: rgb(23,43,77);text-decoration: none;">Complete Dataset - Overlapping view:</span></p><p><span class="confluence-embedded-file-wrapper confluence-embedded-manual-size"></span></p><h2 id="DataWrangling(HighRes)-NotableAspects">Notable Aspects</h2><p>The images are geoaligned but do not line up properly, as shown in the following image:</p><p><span class="confluence-embedded-file-wrapper confluence-embedded-manual-size"></span></p><p>In order to overcome this, a preprocessing step which includes image alignment utilizing open-CV is being investigated. <a href="https://www.learnopencv.com/image-alignment-ecc-in-opencv-c-python/" class="external-link" rel="nofollow">https://www.learnopencv.com/image-alignment-ecc-in-opencv-c-python/</a></p><p><br/></p><p><br/></p><p>Additional access to gbdx systems has listed HR images for the following dates:</p><ul><li>'2017-01-14T02:35:18.346Z'</li><li>'2017-01-15T02:50:50.846Z'</li><li>'2017-11-22T02:50:45.000Z'</li><li>'2017-03-31T03:00:47.616Z'</li><li>'2017-01-14T02:36:11.597Z'</li><li>'2017-01-15T02:51:48.945Z'</li><li>'2016-01-06T02:45:55.697Z'</li><li>'2016-11-02T02:55:58.829Z'</li><li>'2016-06-25T02:47:48.409Z'</li><li>'2015-04-21T02:29:34.173Z'</li><li>'2015-04-21T02:28:29.048Z'</li></ul>

Cloud Providers

2018-08-30T14:51:03Z

A1669892: Created page with " <p class="auto-cursor-target">Approach following companies regarding sponsorship for our project (need to be GPU based)</p><ul><li>AWS*</li><li>Microsoft Azure (Get 100 modul..."

<p class="auto-cursor-target">Approach following companies regarding sponsorship for our project (need to be GPU based)</p><ul><li>AWS*</li><li>Microsoft Azure (Get 100 modules per experiment with student account)</li><li>IBM</li><li>Google cloud</li><li>Floydhub</li></ul><p><span style="color: rgb(255,0,0);"><strong>Update 21/06/18</strong></span></p><p>Google Cloud Engine (GCE) provides $400AUD of free credit (~166 hours for 2xP100's + 4vCPU's) upon initial registration with a gmail account. These credits can be used towards any piece of system hardware, including up to 4xP100 GPU systems. This requires a Debit/Credit card to be added for identification purposes, <strong>but</strong> this card can be used <em>multiple</em> times across multiple accounts with each account receiving the same credit. Hence, with the use across currently owned gmail accounts this should provide sufficient compute power/time for training <em>without being charged</em>. This is a temporary solution though until Phoenix is up and running.</p><h2 id="CloudProviders-AWSEmail-Sebastien">AWS Email - Sebastien</h2><div class="table-wrap"><table class="wrapped confluenceTable"><colgroup><col/></colgroup><tbody><tr><td class="confluenceTd"><p>Dear Edward,</p><p>We have been provided your email address by Sebastien Wong who is currently acting as supervisor for my honours project. Sebastien mentioned that you may be able to assist us in our request for GPU-based computation needed to successfully complete our project. The honours project is a required research project for our final year of Electrical and Electronic Engineering and our chosen topic is in machine learning. The project objective is to predict changes in high-resolution spatial data using sparse low-resolution spatio-temporal data. We are hoping to publish a paper about our approach, and would be happy to answer any questions you may have.<br/> <br/> As mentioned, we are looking for cloud providers to sponsor us with GPU enabled instances for use in our research and would be happy to provide acknowledgement in our paper and presentations. Ideally, we are looking to run a GPU VM intermittently for next few months, moving into more full-time usage mid-year. The project concludes at the end of November with a presentation of our final research. Consilium Technology is already sponsoring us by providing access to high resolution satellite images for us to use as our data set. Please let me know if you have any further questions regarding our project or request for resources.<br/> <br/> Kind regards,</p><p>Liam and Lucas.<br/><br/></p></td></tr></tbody></table></div><h2 id="CloudProviders-EmailTemplate">Email Template</h2><div class="table-wrap"><table class="wrapped confluenceTable"><colgroup><col/></colgroup><tbody><tr><td class="confluenceTd"><span style="color: rgb(23,43,77);text-decoration: none;">Dear {name}<br/><br/>As part of our final year of Electrical and Electronic Engineering at the University of Adelaide, students are required to conduct an honours research project. My group has chosen a topic in machine learning, specifically predicting changes in high-resolution spacial data using sparse low-resolution spatio-temporal data. We are hoping to publish a paper about our approach, and would be happy to answer any questions you may have.<br/>We are looking for cloud providers to sponsor us with GPU enabled instances for use in our research and would be happy to provide acknowledgement in our paper and presentations. Ideally we are looking to run a GPU VM intermittently for the next few months, moving into more full time usage mid-year. The project ends at the end of November. A local company called Consilium Technology is already sponsoring us by providing access to high resolution satellite images for us to use as our dataset.<br/><br/>Thanks,<br/>{name}</span></td></tr></tbody></table></div>

Infrastructure

2018-08-30T14:47:06Z

A1669892:

<p> This section includes work that needs to be done as prerequisite for the main problem, for example defining and procuring data, and setting up a repeatable execution environment.</p>

<ul>
<li>[[Cloud Providers]]</li>
<li>[[Data Wrangling (High Res)]]</li>
<li>[[Data Wrangling (Low Res)]]</li>
<li>[[Docker Image]]</li>
<li>[[HR Image Registration]]</li>
<li>[[Geo-spatial Tiling]]</li>
<li>[[Hackathon]]</li>
</ul>

Infrastructure

2018-08-30T14:44:08Z

A1669892: Created page with "<p> This section includes work that needs to be done as prerequisite for the main problem, for example defining and procuring data, and setting up a repeatable execution envir..."

<p> This section includes work that needs to be done as prerequisite for the main problem, for example defining and procuring data, and setting up a repeatable execution environment.</p>

<ul>
<li>Cloud Providers</li>
<li>Data Wrangling (High Res)</li>
<li>Data Wrangling (Low Res)</li>
<li>Docker Image</li>
<li>HR Image Registration</li>
<li>Geo-spatial Tiling</li>
<li>Hackathon</li>
</ul>

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:41:58Z

A1669892:

'''Supervisor: ''' Brian Ng |
'''Students: ''' Liam Mellor, Lucas Sargent |
'''Consilium PoC: ''' Anthony Milton
----

'''Description: '''
<p>
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.
</p>

----
'''Pages''' <p>
<ul>
<li>[[Operational]]</li>
<li>[[Infrastructure]]</li>
<li>[[Research]]</li>
<li>[[Problem Space]]</li>
</ul>

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:41:46Z

A1669892:

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:41:21Z

A1669892:

'''Supervisor: ''' Brian Ng |
'''Students: ''' Liam Mellor, Lucas Sargent |
'''Consilium PoC: ''' Anthony Milton
----

'''Description: '''
<p>
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.
</p>

----
'''Pages''' <p>
<ul>
<li>[[Operational]]</li>
[[Infrastructure]]
[[Research]]
[[Problem Space]]
</ul>

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:40:29Z

A1669892:

'''Supervisor: ''' Brian Ng |
'''Students: ''' Liam Mellor, Lucas Sargent |
'''Consilium PoC: ''' Anthony Milton
----

'''Description: '''
<p>
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.
</p>

----
'''Pages''' <p>
[[Operational]]
[[Infrastructure]]
[[Research]]
[[Problem Space]]
</p>

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:39:37Z

A1669892:

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:39:00Z

A1669892:

''''Bold text'''''Supervisor: ''' Brian Ng |
'''Students: ''' Liam Mellor, Lucas Sargent |
'''Consilium PoC: ''' Anthony Milton
----

'''Description: '''
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.

----
'''Pages''' <p>
[[Operational]]
</p>

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-30T14:36:31Z

A1669892:

Operational

2018-08-30T14:34:30Z

A1669892: Created page with "This section includes work operational to the project, for example time logging and meeting notes. * Meetings * Proposal Seminar * Time Log"

This section includes work operational to the project, for example time logging and meeting notes.

* Meetings
* Proposal Seminar
* Time Log

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-29T04:36:08Z

A1669892:

'''Supervisor: ''' Brian Ng |
'''Students: ''' Liam Mellor, Lucas Sargent |
'''Consilium PoC: ''' Anthony Milton
----

'''Description: '''
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-29T04:35:37Z

A1669892:

'''Supervisor: ''' Brian Ng
'''Students: ''' Liam Mellor, Lucas Sargent
'''Consilium PoC: ''' Anthony Milton
----

'''Description: '''
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.

Projects:2018s1-109 High-Resolution Change Prediction using Sparse Spatio-temporal Data

2018-08-29T04:35:08Z

A1669892: Created page with "'''Supervisor: ''' Brian Ng '''Students: ''' Liam Mellor, Lucas Sargent '''Consilium PoC: ''' Anthony Milton '''Description: ''' This project aims to develop a machine learnin..."

'''Supervisor: ''' Brian Ng
'''Students: ''' Liam Mellor, Lucas Sargent
'''Consilium PoC: ''' Anthony Milton
'''Description: '''
This project aims to develop a machine learning system capable of predicting changes in high resolution satellite imagery. A convolutional neural network can learn the spatial transforms undergone by a series of images regardless of size, this project will focus on applying transforms learnt in one resolution to images of a higher resolution. Starting with a sequence of lower resolution images taken at various points in time, and given an initial high resolution image, the system should be able to generate a higher resolution counterpart to match any time spanned by the lower resolution series. The lower resolution series needs to be considered sparse as a variety of factors including exact satellite position and cloud coverage will affect the information available for the network to learn. The application of this project is to increase the frequency of available high resolution satellite images, however the approach taken and techniques learned could be applied to a variety of lossy data problems, such as video compression or small sensor amalgamation. This project requires students to: gain knowledge in deep and convolution machine learning technologies, develop and adapt machine learning techniques for satellite imagery (GIS data) and build robust machine learning networks capable of operating with minimal homogeneous data.