Accuracy Assessment of a Land Cover Classification

[Music] hi under the continue and you're watching fringe is remote sensing this video tutorial is about the accuracy assessment of a consideration particularly using the new tools of the semi automatic classification plug-in version 6.4 we're going to use the stratified random sampling for selecting random points that will be for the interpretive and then we are going to calculate the error of the emulsification using the youngest aromatics so let's zoom in so the first step is simple design first we are going to add the official roster to t.j.s here at the roster and select the classification here open here we have the classification roster with four classes water with tap vegetation their soil for simple design we are going to calculate the area proportion of each class so first open the SCP menu the post-processing tools we select classification report here click we open the tool we click the refresh button here and we select the classification raster then we click run then we save the classification reboard text file here and we have here for each class the pixel Sam and the class percentage this is used in this expression where n is the number of samples and we calculate the summation of the difference between the malaria proportion of each class and the standard deviation of each class divided by the expected standard deviation of the overall accuracy all to the power of two please refer to the text of the tutorial for the details and after the calculation we have a total of 567 samples that we are going to certify for each class in this tutorial they are also dividing by 10 the number of samples for simplifying the process the next step of the sample collection we need to open the basket to here for instance from the SP menu in the pencil tool we'll click the refresh button and we select the classification so we add the classification to the first band set here we don't need to change any parameter we just need this been set for setting the reference system that will be used for the recreation we also need to create a new training input here click this button we enter forest and sample here we have the path of the Eternity input where each sample will be saved now we can open the SP - for the basic tool multiple recreation here we have several options the number of points sample we have also other options and set grid minimum distance of samples but the option that we are going to use with this tutorial is this one stratified for the values so they are going to check this and we enter here the expression that allows us to select raster values but first we need to set these settings here the working toolbar is the ROI settings for the minimum and maximum area because we are going to collect 1 pixel size samples so we enter your one for the minimum area and one for the maximum area this way this table will be automatically filled with one in the menu and one for the maximum field now we enter the number of points for the first class 7 and we change the expression so raster that will equal a 1 and we click create points now here we have the first list of samples we can note one for minimum and one for maximum fields according to these settings now we can enter the number of samples of the second class 18 we change the expression accordingly we click create points here the list is updated now we enter the number of samples of the third class 35 and we change the expression and click create points there here we have the list of points now for the fourth class we have seven samples we click create points here we have the total 57 samples with coordinates that would be saved in the training boot but first we uncheck this option calculate signature because we don't need the spectral signature of these samples and now we click run to start the creation here we have the creation samples after a while all the samples will be saved in the training input here we can note the type r which means that only the polygon is created for each ROI and we can note the macro class set D and class ad automatically filled with one we of course need to change these values the macro Cassidy and the facility according to for interpretation if we double click over any of these sample we can zoom ultimately to the sample and of course we need to start the photo interpretation for instance logging OpenStreetMap or other high resolution data here we have the for instance this sample over an urban area and we can proceed with the photo interpretation of course we can also load heigl resolution images here we have a sentinel - image from Copernicus if we double click over the first sample we can see with this false-color composite where digitation appear red here with this dark area which is the water so the first sample is water and we can enter one here it's already one well we should enter the class if we zoom over the second sample here we can see the second sample cover an area of vegetation which is red with this color composite so we need to change the macro class ID here and we can enter the code of the class vegetation which is 3 we can also set the class ID for reference so now we proceed with the photo interpretation for instance this sample cover a road night way we can also see double suite map some can enter here the class - for built up and we proceed for the photo interpretation of all the sample after the photo interpretation of all the samples we can proceed with the actress assessment open the SCP menu here the accuracy to was processing now we need to select the classification to assess cific ation and the reference vector which is the sample the training we set here in a capacity as vector field of the training input and then we click run and we select where to save the accuracy raster and statistics after a while the calculation we can see here the error raster accuracy raster where each value represents a combination between reference and classification here and these combinations are listed here in the table where you can see their raw matrix code the reference table and the classified code then we have the error matrix where we have the pixel count between classification reference here and then we have the very useful area based error matrix please refer to the text of the tutorial for the details with these error matrix we have the area proportion the MDS the area proportional of each class here we have the area from the classification the class proportion here we have the best area estimate where you can see that is different from the raster classification estimate with this area we are also the standard error here the standard error area and the 95% confidence interval of the area so this is very useful for estimating the true the area of each class then we have the producer accuracy the users accuracy and the other statistics for each class of course their overall accuracy and the kapa have justification all these are calculated with ambiences error matrix these statistics are also saved inside this text file here the accuracy CSV file we can open it with a spreadsheet this is our text file separated by a tab so we can use the program to separate the fields here and we can use all these statistics in another program so calculate the other statistics so this was the accurate assessment of a classification if you have comments or questions please join the Facebook group of the same [Music]

Channel: Luca Congedo

Views: 20,656

Rating: 4.9242425 out of 5

Keywords: QGIS, Supervised Classification, Land Cover

Id: H1cL0yhIygg

Length: 11min 54sec (714 seconds)

Published: Sun Sep 01 2019