Deep Neural Network for Crop Classification using Multitemporal Data for Sialkot

Abstract

Agriculture has always been the most important and the ancient occupation of time throughout the whole world. For the survival of the humankind food is the most important requirement and hence the cultivation. Man cultivated different types of crops on the same soil which has been seen effecting the recharge of land resource. Worldwide research shows, there is a proof of increasing food supply faster than the population growth. In Pakistan still traditional approach for crop yield estimation is used which is based on crop cutting survey that led to be tedious, time consuming and statistically disseminated even at macro level. Advancements have been made in this area too using Remote Sensing imagery and Geographical Information System (GIS) for crop type classification and area calculation for the estimation of crop yield. The main objective of this study was to make this process more effective and efficient applying modern approach i.e., deep learning. Four different techniques were applied in the study to achieve an overall accuracy of 87% with Satellite UNET which used image segmentation technique and an encoder decoder network architecture. This approach can be helpful to create a tool that can estimate real time yield of crops in the Pakistan using freely available remotely sensed data for timely management of food crisis and further decision makings. The results showed more accuracy as compared to the traditional tedious methods and led decision making quicker and more effective.