Machine Learning Techniques in Land Cover Classification using Remote Sensing Data

Replacement of traditional farming by modern tools and techniques is the need of the hour. To orient Indian agriculture system towards technology, a support system is required to give a better economy growth of the country. Hyper spectral data collected from a device can be analyzed on periodic basis by exploring the huge data collected in the area of interest. Various remote sensing satellite data in hyperspectral bands can be an advantage in monitoring very tiny variations in soil, land, water etc. Remote sensing data are often degraded by many issues that may include the failure of on-board hardware, signal downlink, atmospheric conditions and overall quality of the sensors in terms of signal noise ratio or sharpness. These factors reduce the quality of the acquired data, making it often difficult or sometimes impossible to extract relevant information. Hence there is need for adaptive algorithms in the areas of data modelling and optimization. Using such data and design of a Deep Learning Network can provide a solution to farming, not only in detecting anomalies but also keep them prepared to handle effectively. This paper makes an attempt to bring an experimental outcome of selection of an efficient algorithm, tool and data to be used for land cover classification. A good spatial and temporal resolution remote sense data, open-source platform, deep learning techniques and different spectral indices recorded by a greater number of bands say hyper spectra range can be a boon to revolutionize farming and alleviate poverty line by improvement in sustainability factor. Satellite imagery converged with digital image processing can be a simple and cost-efficient solution for scientists, researchers, and developers to detect changes, map trends, and quantify differences on the Earth’s surface.