CROPS CLASSIFICATION and YIELD ESTIMATION USING SPATIAL AND SPECTRALFEATURES FROM REMOTE SENSING DATA

Abstract

Absence of advanced technological availability in the Ethiopia Agricultural sector. The estimationof average yield is done by the general crop estimation survey (GCES) technique which relies onthe experiments designed at the time of harvesting period. In the current scenario, getting organizedinformation about the crop status such as crop health, crop growth, acreages, and their estimatedyields were difficult for many developing countries. In addition, the process of crop monitoringsystem was highly prone to error and results biased farming field survey outcome. In this regard, theprocessing and analyzing field survey data was a time-consuming process. To handle theaforementioned pitfalls, we proposed land use land-cover classification and yield estimationalgorithms using a hyperspectral image data. To conduct the experiment, we have utilizedMaximum likelihood, Random Forest and Support Vector machine learning algorithms. Threedifferent Landsat image data have been acquired from the same area in a different time tounderstand the growth status of crops and their pattern in the specified area. A relevant spectralsignatures has been extracted from the LandSat image data to conduct the experiment. From theexperiment result, MLE, RF and SVM classified the land cover with the accuracy of 93, 98, and 97respectively. From the experimental results, we concluded that the application of remote sensingtechnology in the agricultural sector has a significant contribution to improve production efficiency,managing and controlling crop growth, monitoring land use land cover, real-time data processing,analyzing and real-time decision making.