This is the code I'm creating for my minor research thesis on land cover map creation.
My supervisors are Dr Charlotte Pelletier and Dr François Petitjean and my research topic is "Classification of high-resolution Landsat-8 time series by using lower resolution reference data." My project entails outlier detection, image processing, supervised machine learning and time series forecasting. The reason I choose this topic is it has application in better agriculture management on which I wish to start an AgriTech company someday.
The objective of this project is to create a high-resolution accurate map of Victoria's vegetation, using the latest generation Landsat-8 satellites at 30-meter resolution. This would serve as the basis for fire-spread models, algae outbreak detection or pollution management. The challenge is to use reference map which is at a lower resolution, where each pixel represents the area of 250m x 250m. Furthermore, this reference map contains many mislabelling errors and unclear pixels that need to be rectified. Consequently, this data set is used to accurately classifying our high-resolution satellite image at 30m spatial resolution.
Click to watch the video below
