Supervised Classification of Sentinel-2 Images

hi I'm Luca congedo and you're watching fringe as the remote sensing this is the second basic video tutorial about the processing of the Sentinel - image from download - classification so first step is to download the Sentinel to images we have these buttons here and also in the recipe menu these functions here we should select download Sentinel 2 and the main interface window will open so this is the interface or downloading central to images we should enter here a username and password according to the service here and at the moment with the SIOP copernicus we should enter guest as the username and guests as password now we need to set the search area in general we could click this button and then with left click and drag click define the upper left and the lower right corner of the area but now enter these values for the upper left longitude enter 12 2004 2 of the lower light longitude 13 or lower right latitude 41 and then click the find button after a few seconds you'll see that the result of the query and will be listed here all the images in this area will be listed and particularly in this tutorial we are going to use a specific image acquired in May 2016 so in day from enter 2016 5 6 and also the same date in the field to this way we are narrowing the search and will get only the images acquired in this date so as you can see these are the granules or tiles of the image acquired and the 6th of May you can see here the zone we should select this zone 32 TQM which is a an area near Rome so click this button to download the image preview then in download options we should uncheck the bands that we are not going to use than one then nine and Benton we should leave check this option because we are going to download only this image with the preview loaded into Jes unshackle so the process images and log band click this button here select directory where images will be downloaded and click OK as you can see the download will start when the download is finished we can see that in the download directory we have all these files jp2 files which are the central to bands now we are going to convert the central to bands to reflectance in particular click this button here the processing or select from the SCP menu / processing Sentinel tool the main interface window will open and you can see here directory containing central 2 bands click this button and select the directory where all the bands were unloaded here the JPG files are listed we should check also apply this 1 atmospheric correction which is a very basic and simple atmospheric correction and uncheck create band set because we're going to define the band set later so now we can click run define an output directory and click OK the conversion to reflectance will start after a few minutes we'll have the convert bands here loaded in QGIS and converted to reflectance so this is a very large area we need to clip the data to our study area so we are going to use to clip multiple images from the pre-processing tool so select clip multiple rosters and in the main interface window we should click this button to refresh the layer list here you have the list of all the Bands load in QGIS in this case we can click select all selected bands which are the bands that will be clipped then you can use this button here and left click right click to define the upper left and the lower right corner so we can define the area that will be used for clipping so all the images will be clipped according to this area also we could change manually so for instance we can enter here manually the upper left and lower right coordinates according to this tutorial and then we click this button run select an output directory and when the clip is completed we can see the de clip bands are allotted in qgis now we are going to create the input of the suffocation which is the band set so click this button band set and in the maining interface we have the list of the all the band's single band roster loaded in QGIS we should check only the clip bands here when we have checked these bands we click this button to add the bands to the band set and here we have the list of the sensitive bands in the band set we should also order the number of the bands for instance where we have this band 8a which should be set after the mandate and before the band 11 we can use this button to move then order the bands next we should set also the center wavelength so in the quick well and settings select central - and as you can see the center wavelength will be set for all the bands here in the input image we can see that the band set is defined in this tutorial we are going to classify these for land cover classes water tap vegetation in their soil similarly to the first video tutorial now we are going to create the training input file so click this button and create a new training file for instance training and here we have the path to the training file this file will contain both the spectral signatures and the right polygons now we can also display a color composite of the image so in the LGB list select horriston to each one which is a natural color avid for raster is automatically created and load in QGIS we can also edit the RGB list and for instance set the color composite 3 7 10 and press Enter here with this color composite and this is particularly useful for highlighting urban areas we can switch between color composite very easily from the LGB list also a new tool is the RGB list which allows for ordering and editing the RGB color composites so for instance click this button to add a car composite and enter 7 3 2 which is the classic near-infrared post color composite we can also delete color composites and we can switch from the LGB list also using the mouse wheel so it is very very quick to change the color conclusion now the classification dot we can see that this option display and DVI allows for displaying the pixel value of the NDVI when we click one of these buttons here for the recreation so for instance here we can see that there is very low values and negative values for water very high values or vegetation we can also use these buttons here to enhance the contrast improve the display of the image actually improving the stretching of the color composite so now we can create the first row and click here for the region growing Roy you can see that a few pixels are not included we can also change the distance value and and we do the ROI we can see here that these pixels are now included next we set the macro class information and class information for water and we change the color the color of the class here for instance blue we can notice that the class ID is automatically increased and also the macro class is displayed in the tab macro classes we should also change these colors here or the macro class water now we're going to create the second row for the class we'll tap again we should use these buttons to announce the display and stretching of the color composite you can see more clearly these buildings with the white rule then we are going to create a row we can increase the distance value in order to create larger rise and also we click different pixels we can see that the shape of the Royal changes or we can create a manual roy including more spectral variability destroy so for instance and here is polygon next we set the Immaculate information and the macro classic default is roy so built up and across formation here we have saved our second row for the class buildings built up change the color and we also change the color from the macro class here now we are going to create the third row you can change the color composite in order to see more clearly built-up areas so for instance here we are going to create we're going to create a new row for rods so this color composites 3 7 10 is a variable for lighting built-up structures and changing the distance values we can adapt our royal in order to select only pixels belonging to rows so when we are satisfied we can save the ROI change the class information the macro class is the same and click Save as you can see the macro classes is not changing because we have added a new row with the same micro class built up we can create another row for another type of built up you can see here these buildings different from the previous ones in this urban area we have buildings and very narrow roads change again the distance parameter so we can create another row here again we change all the de Cassidy and the class information so here cross information buildings and roads very narrow roads we select the color for the class you can see that if we change the macro class also the macro class information and changes according to the definition of the macro class of course only for macro classes already saved we should collect several rows for each macro class and here you can see the process of collecting several rise for the class built-up vegetation and bare soil of course it is important that these roy's are quite large but also quite homogeneous because we are going to use the maximum likelihood algorithm here after the creation of several rise we can see that the right part of the image we have clouds we could mask clouds in several ways for instance we can create it rise for clouds so using region growing change the distance parameter and set the macro class ad 0 which is a special code for unclassified pixel so here macro class information we can set and classified and transformation cloud it is important to set a unique classy D like or the other Royce and we quick sale we can see that this macro class 0 has been added and we can change also the color here for instance black pixel belonging to this macro class will be set as unclassified after the collection of several rise you can create our classification preview you can also set previous sides of 500 pixels and after clicking this button here clicking any pixel of the image classification algorithm select the maximum likelihood algorithm that we are going to use in this tutorial click any pixel of the image it is important that this process is a iterative process and you can see that the classification preview for the classes has been calculated click this button to zoom to the preview we can see you can also use this radio button to show and hide the preview use the transparency tool so we can see that preview has been classified according to the class ID a classification algorithm we check a macro cassadee we are going to use the macro class called macro class so we click this button we redo the preview and we can see that this classification is performed with the macro class ad so it's quite different from the previous one of the colors and now we can create a preview here and you can see that also we create a preview where there are clouds these clouds are unclassified we have created the previews we should also assess the spectral signatures which is important and in order to perform a good classification so highlight the spectral signatures and click this button to display the spectral signature plot here we have the plot we can also move with the left click and set the plot we can compare this spectral signature of vegetation crop with the spectral signature of built up can show and hide spectral signatures using these check box here we can see the value range which is the minimum and maximum values of the RAI you can also see the signature details here with the oldest information and the sides in pixels the number of pixels of the ROI here we have the values the mean values and the standard deviation of each ROI which is of course related to the spectral signature it is important that spectral signatures are distant in order to avoid classification errors if we click this button here calculate spectral distances we can calculate the spectral distances between each pair of roy's or spectral signatures here we have the surface map to see the distance spectral angle freedom distance and break curative similarity these are very important to assess the similarity of spectral signatures we can also assess the classification disabling one or more spectral signatures so for instance we can uncheck these spectral signatures and redo the preview we can see that this preview has being calculated without the spectral signatures that we haven't checked we could also use this button to delete one or more spectral signatures that we have highlighted but now we are going to select again the spectral signatures and redo the preview spectral signatures are saved in the training input file every time we click the Save button the QGIS Save button which also saves the project so we should click Save quite often now have collected several rice of course more is the better it is important that every class is represented we can double-click the rice list to show resume and directly to every ROI we want to display we can use this button to zoom to the image extent so we are going to create a classification using the macro pass ad and the maximum light code algorithm but or performing the classification it is important to set this parameter in the tool settings of the plugin in particular the available Ram parameter it is important of course to increase this value and you should set a value which is a half of the computer available Ram so for instance we set one gigabyte of available Ram then in classification output we can click run select an output destination after a few minutes the entire input the image would be classified if we zoom here over the cloud we can see that the cloud area has been lived and classified of course we could also in this mask of cloud cover with other methods that we are going to see in other tutorials so we can see here the classification of built-up areas and also we have some classification errors resistance here of airplane soil we have also other tools for improving this classification using this post processing tools for instance the tool added raster for editing the raster values directly and also classification sieve for removing small patches classification erosion and classification relation but we are going to see these tools in another tutorials so well done we have completed the second basic tutorial and the next few weeks I'm going to upload the several new tutorials about this new version files image magnification 13 and please join the Facebook group and the Google+ community and thank you for watching

Channel: Luca Congedo

Views: 90,753

Rating: 4.9193277 out of 5

Keywords: QGIS, Supervised Classification, Land Cover, Sentinel-2

Id: FcETq8OWM0k

Length: 29min 7sec (1747 seconds)

Published: Wed Sep 07 2016