There has been an extensive use of Unmanned Aerial Vehicles in search and rescue missions to distribute first aid kits and food packets. It is important that these UAVs are able to identify and distinguish the markers from one another for effective distribution. One of the common ways to mark the locations is via the use of characters superimposed on shapes of various colors which gives rise to wide variety of markers based on combination of different shapes, characters, and their respective colors. In this paper, we propose an object detection and classification pipeline which prevents false positives and minimizes misclassification of alphanumeric characters and shapes in aerial images. Our method makes use of traditional computer vision techniques and unsupervised machine learning methods for identifying region proposals, segmenting the image targets and removing false positives. We make use of a computationally light model for classification, making it easy to be deployed on any aerial vehicle.