Land use and Land cover classification using Random forest machine learning in Google Earth Engine

okay so hi and welcome can you hear me yes perfect so basically today is our sixth number of tasks and today we will try to discuss about that land use land cover classification how we can easily news land cover classification using the different types of machine learning method uh we write this question about that from beginners to advanced level we try to discuss all of this missionary algorithms such as random versus then um classification and education please then support with permission and other types of machine learning algorithm how we can use and then make the values and our classification after that we also try to discuss about that how we can calculated that area how much build a video how much the application how much photo plan how much settlement is available for this area so this type of calculation is because and finally we also try to discuss about the CSS unit how can you really make the records assessment using this platform you can also make the what is so let's go let's finish [Music] it's the now can you see image screen yes okay so let's okay so basically today is our sixth number of class and today I will discuss about that how we can easily make the plan used and proper classification using the machine learning and how to make The Landings and finally how to add the least end on your map okay as well as we also try to discussed about that how we will check the accuracy assessment of this language what classification using the Kappa as well as also our liquidity and producer accuracy consumer accuracy so all of those details we try to populate it in the solvent training program so let's go so first of all when you want to make the land use land cover classification map using Google app engine platform we have to follow some step or process and this process is the same for any kind of plan use when proper classification first of all you have to selecting your region of Interest which area you want to make the land use learn about classification map just simply select this area this is your Roi if you want you can import the boundary shape file or if you want you can import the polygon anything in as the origin of Interest after that you have to choose and filter in the image collection if you want you can use the land set emails or if you want you can use the Sentinel imagery okay you can simply choose and then filter this image collection according to the your original application is according to the specific time period according to the specific method okay all of those filtering you apply after that you need to collect the training point you can take the tailing point for water body you can take the training point for presentation you can take the training point for build a failure you can take the training point for proof plan all of those classes with the one who classify just simply take the training point from this satellite images after that you need to merge all of those scaling points together okay so all of the staining point you can simply match and store it in a one variable after that you need to make the training data set so mainly this training data set created by the pixel value band pixel value so it extract the band pixel below and create the training data set this data set we need to divide into two ways some data set we are using for train our model and some data set we are using for testing our model okay so this is that has to be divided after that we need to create the classifier this classifier can be a supported with your machine as well as also random Forest it can be classification relation please any kind of classifier we create and classify the illness as well as also create the classified emails and display the result okay as well also print the confusion Matrix foreign heater and try to show you this part so let's talk so just open the code editor open okay so it open so in this time first of all I can simply search here suppose my reasonabilities is that suppose I want to work for the power of copperia just when I simply find out the reason what is the name about that please search okay so basically I want to work with this region okay just a minute I take the phone and then continue just a minute [Music] okay so let's go so this is my relation I want to work with for this in this time I simply solve this region in here [Music] foreign [Music] [Music] [Music] suppose this is my history area um this is my study area so in this time if you want you can simply import your boundary shift file from the asset app or if you want to just simply grow this region of Internet suppose I simply grow this methodology if I simply the drug and this is interesting I need this map for this region okay so I simply draw this region of the internet look like this about that coil mining region so basically this is the coil mining region so I simply draw okay so now uh isolate my developing guys so this is my result of interest in this time for this region I want to create the land use Land Rover classification foreign [Music] [Music] [Music] okay so okay [Music] okay yeah it's good okay good thank you yes okay so just I think it is made a lot of interest I put the name Roi so this is my first step I done so now I want to uh choose I'll be trying to decide [Music] you haven't shared of the screen [Music] okay choose and filtering image collection so okay so in this time I want to work with for the Sentinel two serialization so for the next simply go to the singularity [Music] okay okay so I imported support business region I simply draw the polygon about my Legion and then I need to choose and filter in this collection I simply open my data catalog and go to the sales collection so here I use these cellular crucial elements you can see the similar to multi spectral instrument in just a minute okay so I simply work with what that you can see I simply go to the top surface deflector thing is I want to use for the create email language classification and then I can simply copy this article snippet from here and then paste it into this code reader okay so now this image collection is important and then I simply give a variable name suppose I put the variable language there is no problem okay so after that this image collection I need to filter okay so for that I simply take a very well suppose image and call this Sentinel okay send email to and paste here Sentinel to user the filter bounce function for intersecting all of those emails according to my region of Interest so Roi can you get the filter audit function suppose I want to make this language Network classification map for Europe 2023 first January 16. [Music] [Applause] okay so in this time I also use here the um filter metadata so I choose only for those emails so this image will be not more than one percent cloud cover okay which will be not more than one percent not covered so use here filter metadata meta return and I simply go to this image properties you can see the image properties just simply cope with our Cloud pixel part sentence isolated at the less than one percent cloud pixel percent so user this less than less than one okay so after that I want to check how many image is available for this region so print is and use other size function okay so now you can see that 21 images I want to bought for this video okay so in this time I want to use the total 21 images and make a medial composite so you get the median median and this median we'll deal with my shape file with ROI ROI so now display this result so for that I can simply [Music] use here map Dot layer and call the image and it will run so now one image is added between this time period we can get this image in here you can see okay so I want to add the visualization so I simply use here this simply click on here and put here suppose before break then green and also blue and custom 98 percent and then so now so now I can get this image look like that you can see okay this time period which is look like this so now from this screen is I want to classify the buildup area water body then proof land and Barrel land okay using the random first classifier so let's say about that how we can easily do that for calculating this type of analysis and filtering the image collection okay so now I also imported this so I simply import that and then simply copy and then based after the object name so now this visualization will be sent with my code okay so now click to the Run [Music] so now I can get this image look like that so now from this image I need to get the training point for a specific class okay so what then what can I do just I simply go to the below and then simply search here suppose I simply uh take a variable okay when you want to take the training point then we can use the different types of spectral or different types of band combination suppose you can see here here I use the true color band composite okay so when I use this another pen combination such as uh you said the linear infinite p h then I put at the Red Band person and then apply then I can get the another types of visualization in this time I can hear the resolution look like that okay so in this time when you are using this type of bell combination then water body look like as a black color okay what our body look like as a black color when you are using this type of man combination okay so in this time I use this well combination and take the training point for waterboarding so click to the new layer and then go to the click in here and then I simply put this name is that suppose water water imported feature collection and property will be I said the property suppose class and set the value for the water jio and then okay so now I simply select the this type of Point okay if you want you can also use the rectangle but it also be here so just simply select that marker and just take the training Point using this way so just by simply take this training point you can see as a um black color when I use this type of pen combination and you can easily take this training Point as a waterboard so I take this tailing point for the water body because it is a supervised machine learning method in the super device machine learning method we need to define the pre-category of okay about this text so I I just simply select that and put this this is the water body foreign [Music] [Music] all frequency this is the waterboarding let us say simply take this training pineapple water board went for another suppose in this time I take the training point for settlement or build up area so what that I use the another one combination for certain element I use that the Red Band then green band and blue band also customer person and then apply so now I find out below here so this is the Builder free as you can see so forth that I've taken a new training point I simply click on the new layer and then click on the geometry and put this name as that build up okay and then import as a new layer feature collection property property name will be same for class and set the value for one okay then okay so now I take the training point so this is the build of area we can easily table in front of the build up area foreign everywhere so we can get this version if you can see the Builder the spinning Point has repeated in here also in here information [Music] so after that I hit the training point for um Maryland okay so mainly parent and we are using at this type of encomposition uh it mainly reflected the short point one so short or minified one is the B 11 and then put it at the near inference okay so this then combination will use for identify the balance as well also Pro plant so just I simply use there and then straight 98 percent airplanes so now we need to find out the balance as well as also build Aquarium I think it will be the red penalty before okay so you can simply go to your website then get all of the combination results so I simply go to this result pen combination Nation combination so in this link from this link we can easily repeat the idea about the pen combination and who is one combination provide with a color we can easily find out from here you can see so I want to use this gun combination uh six five two okay six is the short wave infrared one they are near infrared and blue okay so for that I simply click on here a short wave infrared one let's take the band combination about a particular so short wave infrared one you can see the short drive impaired one is that yeah after that uh we also need to check about this band near infinite is that foreign [Music] yeah so now we can easily find out the panel length so parallel line mainly this type of color identified [Music] so industry simply click on here and then put here such as the parallel ogram parallel is also fellow land so feature collection property will be class foreign [Music] foreign [Music] [Music] and then using the problem on this region feature Collision set the property as the class I'll put here this uh three okay so you can see the sequence zero one two three so in this time we can easily find out the Pro Plan look like this color Okay so this type of is the problem okay it will be problem you can see the step up a light green color which identify the problem and the green color identify the vegetation okay so you can see this is a problem okay so I don't interpret it so just I simply click on here remove okay so this is the problem this type of color ID independent Pro plant also this type of problem okay if you want to check it from the seller ID let's click on here and then we can also get the Google satellite images and also identify the this type of analysis and see yeah so it open so now you can simply yeah so which type of position in this time it's showing the barrel length but uh for in this time period we can get the some uh problem area or here or you can see the problem okay so part of this time we are just taking this group plan from here as a agriculture field from here okay so let's go for that I can simply click the Pro Plan and then click on this smart card and set the problem and think this is going for the program so this is a Pro Plan foreign [Music] in here and get some code line here okay so now after the problem we take the training point for vegetation okay without problems with the vegetation property will be class and put it at the zero four okay so you can see zero one two three and four so there is a lot of presentation you can find out from here you can see so this is a resignation so it's also visitation without implementation documentation [Music] right so now my training point will be done you can see here I add this always training Point such as water buildup barrel length problem and visitation so now all of this training point I need to merge so for that I need to margin so much all of training points together so for that take a variable suppose a sample and then I simply match all of the spinning points together so I put here this sample look like that so I put here [Music] button and with margins field up [Music] on and Martin presentation okay so all of the city I want to just simply March and give me one variable okay so just simply print that's variable sample and click on so now we can find out this place where you can see how much point I call it I call it here total 296. released okay as well as if you want to check what is this telling Point what is that class CNC this class is that we can get the class for this properties is that uh zero okay so zero means that water so for the this tailing point I use for water okay so all of the steering point is the level okay their level is coming in here you can see for the 79 daily point is for the class one so here I can get all of those things one by one so here you can see I can get this result in here plus one area class as well as also you can see all of those class all of those class training Point as a level so this is my third step which I show already in the group you can see I simply select Merida Bill various then choose the Sentinel two satellite image and filtering this image collection is also done I take the cleaning file okay it's also done I merge the training point it also done so now I just simply make a training data set [Music] so this type of band I need to store it any variable okay so who is where I use so for that I can simply recovery about suppose bends dance and then make a list till this time all of those band lists are included here here you can see here I can get the pen so I use only for the tail meter band okay there's no problem so 20 meter 20 meter then I will be using here so for that you can see 10 meter back we can get in here only for the B2 B3 B4 B there'll also be 5 is the 20 meter V6 B7 B8 okay and then also you will get the P11 so I simply write all of those bands so I put here go to my code in here and both had that Bell is that B2 I use then also used at the B3 then you get the B4 then you'll get the B5 B6 [Music] use at the B7 after that V7 we use at the b h is [Music] and after that we are using at this we get after that we already had that only for the 11 it can be 12 so be 11 be b11 and P 10 okay so for that I simply select my all of those band so now all of this band pixel value I want to subtract all of those training Point suppose uh this tailing point I extract all of those benefits all of those benefits so all of this training coil is track all of those and basically this is okay so this is my training support then what can I do for making the training get message so basically it's called the Ben collection and then we also will be using this making data set by bands feel very [Music] good so now I seem to take a very well suppose training and now I simply call my emails so this is my emails you can see I already create this image it will be stored in this variable previous simple follow this variable name can paste here after that you will have the select function and from this image I select all of those bands are included in here okay paste here after that I use the function this function call that sample region okay I'll put all of this argument one by one basically this sample region function helps you to collect the training to easel value from the band so either the sample regions and make a dictionary after that I simply make this collection so collection is that your trailing sample or training Point all of the straining point is stored in this variable sample so simply copy that and use here okay and then properties property will be same for all of this class you can see when I take the water or I use the same property class okay property name is the same all behind so for that I use here this properties name is you can see you can see here so here we can give all of this principle you can see it is suppose so you can see the properties all of those band value we can get all of those band value in here okay for this on any point also you can get this all of the straining point you can see all of them okay inside selection here so basically this is my training data set okay so this training data set I want to use in this time so this training data set for training in data set divided into two way one I use for train my model train my model and another I use for test my model paste my model for accuracy accuracy and create my model I use for so here I use here this 80 percent data for really and for this data I use here this twenty percent data for testing okay if you want you can use that the 70 and 30 there is no problem so now I want to use the ratio 50 and 20 percent and take it very well suppose okay so I simply copy the train I'll paste here tell us the function random column and basically I use here this random column function because I use at the 80 percent data will also take the 20 percent data randomly okay randomly I select that okay for that I use at the random column so now all of this data is stored in here so now from this data set I have to divide so I take a variable suppose both the variable name is that brain don't say I put this here I store the 80 partial data so simply call the data set values are the filter [Music] e Dot filter less than equal okay that's a single function about that so I use their function uh e dot filter to replace them less than less than put here this random random Point okay after that they say we put here data set dot filter art engine filter then prepare the greater then or equals 0.3 okay so now you divided that 50 percent less than 20 percent data and square it in this variable and also studying this variable so now I create my data set so this data set I use here empty percent data and twenty percent data so now I need to make the model okay so I'll create the model water body as a blue color is same color you can get in here you can see the click on here we can get the chat look like that so this is the position five thousand one square hectare also you can see the Brooklyn 1560 Newton also you can see the barren land is the 2949 hectare then build up 700 acre also water body only 100 from here and then missed from this code I just simply copy the listen code because listen code is the same for all time okay so it will be start from here say position of the living and then I'll say yeah okay let me copy and then simply I paste here I paste from here in this time I simply change something about that suppose I put this uh okay I put this title name is that okay so okay plan is okay just actually put it at this my new region name of foreign after that we also need to add here after that I simply put here in this time I put out our name so only the name so first of all what I go to it then build up area then we are using at the balance so you get this remove that so then after the parallel so put here at this problem [Music] simply starts here alone take them opening now and paste here I want to show the group uh so I simply put at the peak or in the code is [Music] important [Music] platform foreign [Music] thank you yes I have a summer question I am written also in the chapter yeah what is the question you have well if it is possible to compute the omission and commission error and then if there is a way to compute the optimal number of training classes directly in Google Earth engine and finally if there is a way to Output the optimal number of trees in a random Forest algorithm in order to perform any classification with different number of trees or automatically directly Google Earth engine and then have printed into the console of the optimal number of trees that we have to set in the classification [Music] phrase [Music] yeah but it's possible to yeah okay but for that we also need to use the ground fruit data okay if you have the ground floor data then we can also we are also using the ground root Point suppose mainly this point we need when you classify the agriculture blend different different agricultural land suppose this is the rice this is the sugar cane this is the mice so this type of analysis we want to do we just take the paining point from the ground okay we can simply take the GPS location and also import this VPS location in our um and make this classification okay so that's why we are when you want to make the individual phrase We want to classified or individual core plan you want to classify then we need to the ground fruit Point okay our ground fruit about the specific for the negative longitude then we are using this data it has a point shape file in Google Organization platform and extract the panel pixel value and make the classification about that so I also tried to show you the next class about that how we can easily use the ground crew data and make the different types of uh classification mainly problem suppose rice sugar can mice so this type of problems how we can easily classified using the ground group data using the same process you can easily apply and make the also classified the different different fields and also classified using the ground okay easily learn about that also okay [Music] [Music] so so okay okay [Music] 24. okay okay it's perfect I will sign up in my calendar okay okay July [Music] okay and this will be the last one yeah yeah okay but you can also get the lifetime teaching support there is no problem after completing the thing so only if there's any problem then we can also just uh for for joining any meeting we can easily connect with the meeting and also talk about the problem okay there is no problem okay thank you thank you so much have a good day okay bye you too bye foreign

Channel: Study Hacks-Institute of GIS & Remote Sensing

Views: 8,140

Keywords: land use and land cover classification using deep learning techniques, land use and land cover classification, google earth engine tutorial for beginners, google earth engine tutorial, google earth engine python, geeta govindam hindi full movie hd

Id: Z-DPRCWWaqg

Length: 60min 58sec (3658 seconds)

Published: Mon Jul 17 2023