Machine Learning in R: Image Classification in land use LULC

Explore what machine learning is, its major algorithm types, and how supervised, semi-supervised, unsupervised, and reinforcement learning are used in remote sensing and geospatial analysis.

Explore image classification in geospatial analysis with supervised and unsupervised approaches, applying algorithms like random forest and SVM to Landsat and Sentinel imagery for land use mapping.

Explore common supervised and unsupervised image classification algorithms, including minimum distance to means, maximum likelihood, decision trees, random forest, and support vector machines, with practical examples and accuracy considerations.

Explore the RStudio interface, including source, console, environment, history, and files panes, and learn to set working directories, access help, and run code across Windows, Mac, and Linux.

Learn to install and manage packages in our studio, loading from repository, CRAN, or package archive data. Load with library and call specific functions for remote sensing analysis.

Learn how to define and assign variables in R using the assignment operator, view results in the environment pane, and perform basic arithmetic with named variables.

Learn how to assign variables in R, compute a sum and a product (x and agric), print results in the console, and name variables meaningfully for clarity.

Explore how R uses objects and symbols to store data, covering atomic vector types (numeric, integer, character, logical, complex) and attributes. Learn to work with factors and levels.

Learn data types in R, including numeric, integer, logical, and character. Use as.integer for integer conversion and explore factor levels with ordering and paste for strings.

Explore vectors and vector operations in R by declaring, indexing, combining, and performing arithmetic, noting type conversions, recycling rules, and built-in functions like maximum and variance.

Learn how factors represent categorical data in R for analysis, with strings converted to factors by default and levels mapping to integers in alphabetical order; use levels to set order.

Learn data frames, the most commonly used two dimensional tabular data structure in R. Inspect their variable types and observations with the structure function and summary.

Explore control structures in R, including if and else statements and vectorized forms, and learn how to print results and apply conditional logic to vectors.

Learn how to implement for loops in R to iterate over index ranges and vector elements, print values, and write generic looping code.

Learn to input data into R by reading Excel, CSV, and text files, set the working directory, specify headers and separators, and preview and summarize data.

Learn how digital satellite images use pixels and multispectral bands, and compare spatial, spectral, radiometric, and temporal resolutions to classify land use and land cover.

Explore passive and active remote sensing sensors, multispectral and hyperspectral imagery, and platform types—from ground to satellite—for land use and land cover classification.

Master the basics of remote sensing for land use and land cover mapping, perform visual and automated change detection, and apply spectral signatures and vegetation indices for classification.

Define and select mutually exclusive, exhaustive land use land cover classes for a supervised workflow, then train with representative spectral statistics to classify pixels into land use types.

Explain Landsat data characteristics, focusing on Landsat 8 with OLI and TIRS sensors, 11 bands at 15–30 m, and level 1 and 2 reflectance products.

Learn to reprocess optical remote sensing data through radiometric and geometric corrections, atmospheric correction, and data subsetting, using Landsat 8 level 1/2 and Sentinel-2 products converted to GeoTIFF.

Explore processing Landsat 8 data in R, unzipping archives, stacking spectral bands into a raster stack, and optional data cleanup and pyramid visualization steps.

Load Landsat image data into RStudio with the raster package, define and verify the file path, then unzip the dataset into a folder named after the file.

Learn to load Landsat raster data into R, select spectral bands, and build a raster stack for visualization and analysis, including creating false color composites and inspecting metadata.

Automate batch processing of Landsat images in R by unzipping, stacking bands, renaming files, and creating pyramids for efficient visualization.

Explore visualizing Landsat images in R by creating single-band and true color composites, cropping and subsetting images, and analyzing histograms of the green band to assess land cover features.

Explore Sentinel-2 sensors, their spectral bands and 10, 20, and 60 meter resolutions, and how level 1C and level 2 products enable radiometric corrections and cloud masking for land cover.

Learn how to generate representative training data for land use land cover classification, ensure independent test data, balance classes, and apply sampling, augmentation, and validation using random forest and SVM.

Prepare training data in R by loading sentinel-2 imagery with seven bands at 30 m resolution and organizing training polygons for random forest and support vector machines classification.

Prepare training data in R by converting the shapefile's class names to numeric factors (1–5), renaming raster bands, and extracting pixel values into a data frame for the classifier.

Visualize spectral signatures in R by aggregating training data to compute mean profiles, plotting profiles across bands, and adding lines and a legend for vegetation, agricultural vegetation, grassland, and forest.

Learn to perform image classification in R using random forest, including data preparation, downsampling of minority classes, training with up to thousands of trees, and evaluating with confusion matrices.

Extrapolate the saved random forest model to the Landsat image, predict pixel classes, and visualize results to assess initial accuracy. Revisit training data if misclassifications appear to improve map quality.

Visualize random forest land-use results in QGIS by adding the raster, classifying values 1–5, and applying custom colors and labels for bare soil, forest, grassland, urban, and water.

Train an svm in r for image classification, balance training data, tune gamma and cost with grid search and cross-validation, then predict and visualize land cover.

Explore accuracy assessment of image classification using confusion matrices, visual and quantitative controls, and key metrics like overall, user, and producer accuracy, with emphasis on reference data sources.

Students validate support vector machines classification using provided code, compute uzak accuracy, producer accuracy, overall accuracy, and the copper coefficient on the same validation dataset used for the random forest.

Create a land use land cover map from your final image classification in QGIS in a print layout, with a legend and north arrow, exporting at 300 ppi.

Explore the instructor's Udemy page and YouTube channel Jio World to discover remote sensing, data science, and machine learning courses, updates, and free content.