Deep Learning based Target detection and Recognition using YOLO V5 algorithms from UAVs surveillance feeds

Abstract

This study aims to show how the most recent object identification algorithms (YOLO V5 and YOLO V7) may be used to recognise targets in a real-world setting using surveillance feed collected by UAVs/Drones.The Russia–Ukraine conflict provided UAV imagery that was gathered and processed to better understand how object identification might result in deducing vital inputs for rapid and timely decision making in a warlike setting.Inference on the video was performed using the best and most accurate models from YOLOV5 which was obtained from open-source UAVs feeds.Due of its superior OD capabilities, unmanned aerial vehicles (UAVs) are essential to the realisation of any robot’s full autonomy.Although the YOLO V5 model was trained for a greater total number of epochs to substantially improve the object class detection accuracy.By analysing the Confusion matrix and the performance hyperparameters of the algorithm, this study compiles the performance metrics to make conclusions and optimise outcomes.