All of them have 30 meters per pixel resolution so we do not need to perform upsampling or downsampling to perform analysis. We will pass more data into the k-means algorithm and visualize scenes in false color so the input dataset consists of bands from 1 to 7. We need for it only band 2, band 3 and band 4 of Landsat 8 scene. It is a good idea to look into a scene before the analysis, and that’s why instead of jumping in the deep end we start with data visualization. You may use any other scene, especially images from your area of living! Data Inspection I’ve used scene LC08_L1TP_018030_20190630_20190706_01_T1 which covers the Great Lakes region (Lake Erie, Lake Ontario, Lake Huron) from which I clipped Toronto (CA) area. To perform this task you need a band set of atmospherically corrected Landsat 8 images. You should install those packages (from the conda-forge channel): We use the conda package manager to create a working environment. You may have data labeled only partially, then use of k-means on the unlabeled data is the fastest method to label unknown values. Clusters will be quite large but there are smaller dots – ships – which can be clipped from the image based on the simple statistics. Imagine scene without many details (like sea surface). Scenario is simple, you have scene without clouds, train on it unsupervised classification algorithm and get multiple labels, then you do the same for future scenes and if labels’ borders are not aligned then probably you get a clouds. Data compression (instead of thousands of different pixel values you get dozens of them, file size may be much smaller).Still, there are areas, where unsupervised classification rocks: Unsupervised algorithms don’t give you that opportunity. In truth, supervised classification gives you better results and you know what you’re classifying. Unsupervised categorization of remotely sensed data seems to be impractical. Why do we perform unsupervised classification of the satellite images?
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |