Smart Agri-Farming on Satellite Imageries using Machine Learning

摘要

This study aims to help farmers by using open source software that employs machine learning and hyperspectral images to analyze farm characteristics, which include crops, soil, and climate. This study makes use of two datasets, i.e., 270100 images from LANDSAT 8 and classified images from MODIS dataset provided by Google Earth Engine to classify land type, which helps in detecting farms in the future. Random forest algorithm was used as a classifier for multiclass hyperspectral data. Training the model acquired an overall accuracy of 0.997 that helped to determine the type of land in a geographical area. This paper conveys the first model built by us from various other models that are planned to develop. The data from our research work is conveyed to a farmer by means of a web application, which is built using a Spring framework, Grafana, JvaScript, and several other web technologies.