Applying different techniques of localization on drone imagery related to agriculture

Abstract

Pakistan's major economy depends on agriculture, but we have seen many problems with automation and technology due to a lack of information and lost a lot of crops. In order to get high yield and more production along with the savage of extra labor time and money, we need automation in this era. Automated and high-resolution imagery (temporal and spatial) acquired from low-cost drones provides an opportunity to push the horizon of precision agriculture especially in weed detection, pest attack detection, and yield estimation. The state of the art approaches involving drones for precision agriculture does not include active inspection of the area from drones. This requires making drones autonomous. This thesis aims at applying deep learning at drone imagery acquired in an inquisitive manner. A lot of methods proposed for the localization in agriculture, in that research we want to find out the more suitable in accordance with our current economic condition. First, we tried to detect weed and localization of agriculture using deep learning techniques, but faced a lot of issues, like dealing with intercropping and a lot of more. Then we moved to some computer vision techniques, like SLAM, which helps to cope with the problems that we faced during deep learning techniques. Then we move to one of the simplest methods of localization in agriculture which requires less change in the field and will give more useful and accurate results of localization. At the last, we did compression and observation on all of the above methods of localization and will tell about their advantages and disadvantages