Land Cover Classification using Satellite Imagery and Machine Learning Algorithm

Remote sensing has evidenced itself as a more efficient source of quantitative data than ground based surveys in terms of time and space. Remote sensing is crucial in the collection of data for many aspects of life or areas, such as political, economic, Urban Planning, Agriculture, Forestry, Barren, and even scientific. Land cover data is useful with regard to planning and decision making processes in urban areas and it has several solutions in terms of mapping and monitoring of the urban areas. A satellite image categorization based mainly on urban planning, land-use mapping, and environmental monitoring is used. Within this context, this study explores the possibility of bringing reinforcement learning in conjunction with pertinent point-matching techniques into the analysis of Landsat satellite picture classification via the Google Earth Engine (GEE). As opposed to the classic application of traditional supervised and unsupervised model in the process of satellite image classification, the RL presents a dynamic approach, strongly dependent on feedback for possible improvement over time in classification accuracy. This is further enhanced by key point matching, which makes it possible to identify important characteristics in many temporal pictures, recognize objects, and detect changes more robustly. In our discussion of a unique framework, we highlight the potential advantages of feature correspondence and adaptive categorization in complicated satellite data by integrating RL and key point matching with GEE's cloud based infrastructure. Results from experiments show how well this hybrid model works to increase classification accuracy.