Accuracy Assessment || Kappa Coefficient || Overall Accuracy || User's Accuracy || Producer Accuracy

[Applause] [Music] hello everyone uh today i am going to make a video on accuracy assessment by using the copac officiant we are going to use copper method for that i am going to open my arcgis over here before that i am going to show you my classification image i have classified my llc in this way you can see it now i'm gonna cancel it first in my set my projection and simply click ok and add my data you can simply see it here and i'm going to set it as default i'm not making any change over here i am going to just level my data first of all it is forest land after this this is vegetation after this this is water or the river body i'm going to give it as river and after this i am going to i have assigned it is sandy area you can simply see it and i have assigned it is agricultural land i will show you the cross reference which is my original file one that i have shown you before now this is settlement after this you can simply click apply and go with okay and as i have i have assigned one is you can see one is forest two is vegetation three is river who is sand area and is agricultural land in the six is uh settlement now uh how we're gonna proceed with the accuracy assessment by the copper to calculate compare coefficient uh you can simply have to cross reference your classified image with your referenced image that means here i am going to take a google earth is a cross reference as i will cross references with the google image so we have to determine the truth or false value of our sample taken from this image classified image so for the sample we have to uh select a certain number of so point sample to proceed so without delaying i am going to create a shapefile of a point over here you can simply see it here i am going in my working folder in geotext studio and in accuracy assessment i am going to create a new save file and i am giving it name as sample point and here i am assigning it as point in projection set your projection after that you can simply click ok and we have to first of all now we have to create our point number of simple over different parts of area and while taking simple we have to take a simple in quotes area mostly dense area don't take it in don't take in it is like this in the very core zone or the uh i mean in the confusing zone you have to take it in the area where the uh this land type has more area this means uh this is vegetation part and i am taking my sample in the poor area that means where the vegetation land has more area i am going to take it there this means here don't take the image in like this here it will make uh difficult for us to cross reference in the google image if uh it get stuck in between the three different type of landings so for that now with this i am gonna start my editing simply start editing and take a sample point over here and with this now i am gonna create a point and i am going to set it first of all in the water bodies my first point has been drawn and then go to uh forest area and if in forest area i am going to the core zone as i have mentioned already we have to take it uh from the core zone so i am taking it here and similarly we have to do uh all over the area over here so i am taking it rapidly from here you can see it here i am taking it [Music] zooming in and taking in i am randomly taking a lot of point over here you can see it here uh now from here i am going to show you how to grow it like this and i am taking number of simple over here [Music] i am going to take a lots of point so that it will make a sense while giving an example to you up and over here too i am taking simple point i am taking simple point from all parts of my area you can see there this is forest john and this is agricultural rent you can simply see it over here and this is the settlement area too you have to take it like this and you can take it from here too and zoom in and take your sample in the course area as i have already mentioned i'm just taking a point over here you can see i have taken number of point over here and you can see how many point i have taken over here you can simply go to attribute table and you can see i have taken nearly about 31 point so i'm gonna make it uh nearly 40 point over here now 14 not 35 i have 31 32 point okay i'm doing with that point no need to add more later on it will be very time consuming to go with this now i'm saving it again just save it and now after saving you can stop editing and now you can make a change over here go to open attribute table and add a number of field over here you can have a user i am going to add a field user because this is user classified image so i am going to see save it as user value is we have assigned this value this is user value from a classified image so i am giving it his user value and saving it as float and simply clicking okay sorry uh i have uh taken a space over here so it is showing here so it is making us done giving us a new name is user and underscore value so i am giving it is s and now over here what we have to do is you can simply see it in which location the this point falls we have to identify and we have to assign it over here now i'm making it little bit small as it is covering whole area of the canvas so i am it makes us difficult to take a hello so i am simply first of all going with that and now you can see it here and i am going with the double one so don't worry over here this is river core area this is river you can see it here so we can here uh make a start editing make a simple point and you can here typer give a user value sorry it is zero now i'm going to give it as uh rebar is three value so i am giving it three and going to second value and you can simply go to like this and zoom to this and you can see this is forest value so and we have to assign it as one you can see and go to third point and zoom to and you can see it is under vegetation so i am making it is two similarly now we have to do it for all our points 32 points over here so i am going with that and this is agricultural land i am giving it as five and i am giving it a 0.4 and zooming it and it is agricultural land again so i am giving it as five you can see it here you can get it accord from here or the id or the user value as i have described it before too zoom in and you can see it is vegetation again and you can go for vegetation and vegetation is number two and good for like this with the forest and forest is number one and go with zoom two and this is vegetation and this is five this is river and this is three number nine this is as reward and go to number here and this is number 10 you know you have to grow one more 11 go for red means sandy area this means we have full value over here zoom it and this is settlement area you can simply make it a 6. like this no need to worry just go on with your classified part and i am doing that too and this is also 6. just here classified image reference many and you can see there 20 this means means agricultural land so i am giving five this is also agricultural land so i am giving five here vegetation and this is this is again vegetation so we [Music] so i am going with four this is to three you can see which value has we have low value oh that means sandy area no no this is four one two three taken less sample from it so we can add our some points again from here we can make 35 or something like that so again after we start editing we can add our point to we have less value over here um you can see this is our value so i'm going with this hmm i'm going right farthest too taking sample from forest period giving some points now i am saving it and after saving i am going to open it again and here this is for sandy area and area you can simply now after this we have to save your work now we have to add another field over here now that field name is here you can go with you have to first of all fill over here is producer producer means our referenced image value that means according to our classified value if uh our classified value match with the reference means we will put the same value over here and if it doesn't match without the reference value uh what the reference land type is we have to assign it as that value this means if our classified uh land is 1 over here and in the reference part it is is vegetation then we have to assign the producer value as refreshed image that means we producer value gives us the reference image land type that is original land type so with the producer value we are gonna check accuracy this is our original land type you can take it is that producer value is the original land type so we have to cross reference and set a new value original value over here over our false value and replace with the truth value and you can simply save it over here now after this you can started it same method and now and over here i am going to open my google earth you can convert this is simple point into kml file or you can directly insert it in the google earth you can see i have opened my arcmap and now i am going to here open my that simple 0.5 and you can see we have to go with id so that we can know which point is what and we can cross reference with the classified image so you can simply cancel and check over here now we have total 38 point now we go with the first point and you can go to minimize this as it is uh zoom it to the area and you cannot this is forest area point number one is first forest area and you can go here and here you can started it and simple point in okay now our point one is forest area and zoom it here and zoom we don't need to zoom it is we have assigned our user value user value one assign was forest so this is true so we are going with this true value and now we are going to check zero point point first so for that we are going to minimize it and going with point zero you can simply going with point zero you can see point zero is over here now you can see this is a reward river area and what we have done [Music] you can see in 0 3 means river it is also true so we are going with okay and we are going with zoom tool this is vegetation and here sorry link number two and what point number two will show us now let it be you can see this is a little bit dense forest area or you can see it here our vegetation area will be like this and this is a forest area so this is false so we have to assign point assign this is forest value we have to assign it as first value that means this is false user value we classified point two is uh vegetation but this is forest so we have to assign it with its original value so like this we have to do for all our point and you can see this is agricultural land agricultural endpoint number three and we have to go for google earth and we have to looking for point number three you can zoom out and search for your point number three and you can look it over here no yes this is a agricultural land you can see it over here so it is a truth value so we are going it with the user value you can simply go with this and this is truth so we are going with original one user value if it is false we have to change it you can zoom it and this is also agricultural land for point number four and over here point number four and this is also agricultural land this is also true you can see it here this is also true so we are going with the user value at the original value now we are looking for this and this is a vegetation value and it is point number five it is a point number five foreign you can see this is vegetation part any small small dense vegetation are present so we will go for vegetation and this is a true so we are going with true and this is vegetation so like this we have to do for all this so it is time consuming i will be here after creating all the producer value over here so for now i am just skipping this part now after assigning all the producer value over here you can just go with like this you can see we have total how many total user in the producer value you can see here we have to fill our area with it now select this total and you can see here 8 out of 38 and total we have it in total user in the producer value matched is also total value is matched so now what we have to do we have to fill up this form now we have assigned our classified image as forest and what is come out is also total also forest so we are going reminding all it is zero 0 0 0 0 0 in total this row total is 8 you can see it here now we are going for visitation one you can see it here now vegetation one we are going to do visitation one now you can see uh six total is a six you can see we have total user value as vegetation is two is six how how many sample point is how many simple point come out to be true is you can see producer value match this value is true two two matched one two two matched one two two matched one three total value are matched on agriculture and 5 means 2 1 value this 5 is assigned we have classified this point simple point wrongly is vegetation instead of agricultural land so we have one producer accuracy for uh basic agricultural land this one one row and this one forest we have assigned vegetation forest land is based vegetation land so we have this types of error two either on it so we are gonna define it in our box you can see vegetation we have wrongly classified vegetation in two places two simple point we have rightly classified with vegetation with vegetation in three places and river zero we haven't make mistake in here and agricultural land is one and settlement is zero and our total user simple is total 6 over vegetation length so you can see it here and again we have to go to here and we have to go for three you can see six out of six all are all classified image is accurate with the reference image so we no need to arrive we have six symbol over here so we have to assign it as sorry go to your save file and classified as forest zero vegetation classified is forest zero river classifies residual total out of total we have no error in the river part zero zero zero in total simple six and now sandy area we have to look for sandy area and you can we have no error in the sandy area too you can see here we have no error over here sandy area you can see we have six out of six in the sandy area too we haven't misclassified it sorry i'm going for a word document over here and we haven't falsely classified forest land is sandy area how many sample point zero how many sample point we classified as a sandy area instead instead of vegetation zero in the vegetation land we were 0 sandy area 6 0 and here also 0 but here 6. now this now we are going for agricultural land and here in agricultural land how many simple point we have taken for agricultural land six simple point and how many come out to be a true value with the difference event one two three four five high value come to be a true value but one comes out to be a false we have classified agricultural land we have classified agricultural solar land but it comes out to be a forest land so we have to give a one simple for a forest area here we have classified uh one simple uh in agricultural land instead of forest area other rest we have classified through so we can go for like this 866 also over here now we are going for settlement or our last classification and here to settlement area how many it comes out to be a true you can see it here we have only two value comes out to be a true you know settlement area so we are going for two and how many in two we have classified forest is uh settlement area two we have classified the vegetation is forest area we have classified vegetation as forest area in two simple you can see two two and one is agricultural land we have to make notice or we have to make pigeons of all these in our table over here you can see two in a vegetation one in a forest and two it in these and not two is in this we have two it in this and one it in this other we haven't misclassified sorry you can see our set over here and with this 2 2 4 5 6 you can simply select now we are going to calculate producer equality this coulomb gives a producer accreditation means how many total value we have achieved in this part total classified total referenced forest area or the forest sample point how many total simple point come out to be a forest area in total point we have taken out of 38 sample we have taken these coulomb give us accurate exact forest area of our sample you can see 8 to 10 11 12 go with this 3 4 5 and this is 6 and this is 6 and this is 5 6 7 and this is 2 now for user accuracy what we have to do is you can see here for overall accuracy we have to give it here total or the overall accuracy over here so here total number of collect correctly classified pixel that will be given by our diagonal value is we can see we have classified forest truly total true value we have classified we haven't replaced it with other land which type forest forest aid vegetation vegetation 3 all these true value will given by the diagonal so i am doing this 8 3 you can simply go with like this 8 plus [Music] 3 plus 6 plus 8 3 6 6 8 3 6 6 5 2 5 2 and our project will be valid if our accuracy is over 80 percent otherwise we have to do it we have to reclassify or the classify reclaim we classified our image again and total simple total number of reference pixel you can see it here how many total number of reference pixel we have and we have to put this value over here how many point we have taken we have taken total point we have taken simple number is 38 now calculate 8 3 6 6 8 11 11 12 23 23 5 28 29 30 30 by 38 gives us uh 78 or 79 percent over here 0.79 and we can take it as our image is classified correctly we don't need to classify it is it it is very close to 80 we can take it as 80 so we don't need to reclassify it again now over here uh into 100 just leave it and just give it in 100 equal to 79 accuracy which is quite good no need to worry now our overall accuracy is this now what do we have to do is here we can write 79 percent which is quite good now we are going for user accuracy and over here you can see user accuracy means this value by this value you can see according to user how much have we have classified total number of classified pixel in that criteria you can see it here 3 by 6 you can see here we have classified forest is eight and we have classified rest as this so we have eight by eight this means we have hundred percent accuracy in a forest classification user accuracy and here vegetation we have classified two we have classified vegetation correctly as three this is correctly classified you can see it over here i'm zooming in a little bit you can see it here correctly classified it means we have correctly classified how many vegetation point correctly three in our user a total number of total number of classified pixel in that criteria we have classified total one two three four five six point so three by six that means fifty percent accuracy three by six and river we have also for river also we have six by six you can see this is also 100 percent in sandy area you can see sandy area six by six this is also hundred percent in agricultural land we have five by six this means uh nearly this is 83 percent 83.83 percent 83.33 percent you can calculate also and now settlement um this settlement means two by six by six means thirty three percent one by three that means thirty three percent this is very low you can see it here is it is very difficult to classify the settlement area so it is giving us that false value and you can see it here for producer accuracy we have to take it total from this one this is correctly classified part so for producer accuracy we have to take only corrected one this is correct and we are taking this this means 8 by 32 producer accuracy 8 by 32 you can see we have 8 by 12 not that means 66.67 percent you can simply go with like the 66.67 percent and you can go with vegetation too similarly hired by no vegetation three by five three by five means three by five that means 30 percent this is a real accuracy is you can understand and river is six by six this is hundred percent and sandy area also hundred percent is we have study 71.42 percent settlement area 2 by classified total 2 by 2 this is also 100 percent 100 percent now we have this value now what we have to do is now we are going to a great copper value which gives us our total accuracy over here and you can simply use this one like this after this yarn represent total number of simple taken that means we have taken 38 point in this value it looks vast numerical uh expression but we don't need to worry this means truth value or the diagonal value you can see i i value that means forest forest value you can see it over here forest forest value that means eight vegetation vegetation value that means i i same same value correspond with the same value that means corresponding the same value it will give sort this this will give us this that means diagonal value summation of this diagonal value sorry summation of this diagonal value and this what this means i plus means this is row total row column column row total into coulomb total i will show you row total this means this row total equal to this root total equal to 8 and this row total equal to this coulomb row total equal to 12 so that means we have to sum uh it is 12 into 8 plus 5 into 6 plus 6 into 6 plus 6 into 6 plus 7 into 6 plus 2 into 6 like this we have to summation like this so i'm gonna show you all this expression over here now i'm copying this expression ctrl c and i am giving it as control b ctrl z ctrl c sorry now over here i am doing calculation as i have mentioned before too yun means 38 38 and we have multiplication shift multiplication 38 into summation of x i that means x i mean summation of all this diagonal value and we have this diagonal value is bracket 8 loss we have used that diagonal value over here you can see i'm gonna copy this for time saving ctrl c and i'm gonna copy paste you can see this is a diagonal value now how what will be its output i will show you over here you can see diagonal value over here i'm taking and showing all the calculation over here so it looks like this now i am using calculator over in the my hand 38 into 8 plus 3 11 that means total it gives 30. 11 12 23 23 7 30 38 into 30 it will gives 11 40. now you can see after this value we have to calculate for this expression as i have already mentioned it is a row into coulomb total of every row in column you can see i will show you over here i'm showing it ctrl v and i am and you can see row into column total means i will show you this eight into two well 8 into 2 well plus 5 into 6 into 6 into 6 into double six into six and seven into six and two into six i am doing first so i have read all it at once five into six plus uh six into six plus eight into six into six again double six into six as i have already mentioned plus and five into six six into seven into six seven into six plus 2 into 6 2 into 6 in total this is the summation of this it is known it is total coulomb total into row total now i am using calculator over here to make it fast and this is total 252 now we can use it as 252 252 simply now copper copper value equal to you can simply see it here upper value equal to this value 1 1 0 0 minus this 2 5 2 by uni squared this means total number of observation we have taken or the simple taken now equal to we have taken 38 simple so it's square you can simply cancel this one and you can write it over here 38 square minus this 252 same value row in the column now coefficient equal to i will show you the value later on and show you little bit explanation on it edit it by edit it by simply using like this for now edited by 38 square minus 38 square minus 252 it will be like one one one one nine two now rotate it by edited by one one nine two you can see overall accuracy this is total 75 percent nearly 75 percent now you can see the change over our overall accuracy and it will appear equal to equal to 0.75 percent then we have to multiply it by 100 so it will give us in percentage after this this is our copper assessment value now you will decide whether you have to reclassify it or you can do it here 75 also a good it is not bad classification as well if you have over 80 percent it will be very good i'm just showing you for you it doesn't matter is we have done very good classification so with this we have done our part from accessing the copper assessment in the overall accuracy and producer accuracy user overall for copper coefficient or total value so now i am showing something slide over here the usefulness of some else classification in overall collect data reference ground truth value in the compare reference to the classified map as i have already mentioned and you can see reference data that means we have taken reference data is google earth and reference data the same thing what types of reference data we have to choose and if that depends uh our result in accuracy we are going to do it by pixel that means we have taken a simple value as pixel simple point we have determined point as pixel and this is our reference data how it looks you know how it we classify and remaining all this thing our sampling method how we will be going we have to take it in the middle part so that it doesn't get randomly confused and it will gives a separate result you can see this will give us less less accuracy or we have we get little bit more confusion but that types of high reference data will be very useful to do our tax while classifying and we are while taking as reference to and this is our example if you need it need this document you can just comment down and this is our total accuracy and one thing uh we can find some different formula over here using by people but i have found this one and this reference is i'm showing it here the book written by i'm showing you over here the book written by this and we can find this here here 568 [Music] you can see it over here i have taken that formula from this books and i have referenced this book so i am going with this no need to worry we have done very good you can see here is a separate example here total simple and its copper assessment is 57 percent which is now less than ours so we can be very sure we have done our part very good so no need to worry with this i'm gonna end my video here thank you keep watching

Channel: GeoTech Studio

Views: 8,987

Rating: 4.9078341 out of 5

Keywords: accuracy assessment in arcgis, accuracy assessment of supervised landsat image classification in arcmap software part 2, accuracy assessment for image classification, accuracy assessment of an image classification in arcmap, kappa coefficient, producer accuracy, kappa accuracy assessment in arcgis, confusion matrix, confusion matrix sensitivity and specificity, kappa coefficient formula in excel, kappa coefficient in arcgis, confusion matrix explained, overall accuracy calculation

Id: w-JMgW_9cz4

Length: 57min 8sec (3428 seconds)

Published: Sat Nov 21 2020