izpis_h1_title_alt

Reinforcement learning for efficient UAV-based computer vision
ID Colnar, Brin (Author), ID Pejović, Veljko (Mentor) More about this mentor... This link opens in a new window, ID Machidon, Alina - Luminita (Comentor)

.pdfPDF - Presentation file, Download (5,30 MB)
MD5: F3015B23B573C2B27982102F90FA6499

Abstract
Weed control in precision agriculture illustrates the broader challenge of optimizing operational efficiency in dynamic environments - a principle relevant to fields as diverse as financial markets and environmental monitoring. To effectively meet these diverse needs, we have developed a suite of versatile algorithms that select the most appropriate machine learning model for weed recognition in aerial images in real-time, based on context and operational constraints as a UAV flies over an agricultural field. Our algorithms dynamically choose from several pruned versions of the U-net neural network—ranging from 25% to 100% of the original model's capacity—balancing accuracy against resource consumption. Our approach has proven effective, matching or surpassing fixed-model methods. For instance, our upper confidence bandit is able to achieve same Intersection over Union (IoU) metric as 25% pruned network but at a 15% weight reduction, optimizing energy use. Our algorithms show adaptability, as they are able to prioritize different operational needs, such as extending drone flight time or maximizing segmentation accuracy, depending on the situation.

Language:English
Keywords:reinforcement learning, unmanned aerial vehicle, precision agriculture, computer vision
Work type:Master's thesis/paper
Organization:FRI - Faculty of Computer and Information Science
Year:2024
PID:20.500.12556/RUL-166126 This link opens in a new window
Publication date in RUL:20.12.2024
Views:25
Downloads:0
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:Slovenian
Title:Spodbujevalno učenje za učinkovit računalniški vid na brezpilotnih letalih
Abstract:
Nadzor plevela v naprednem kmetijstvu ponazarja širši izziv optimizacije učinkovitosti delovanja v dinamičnih okoljih - načelo, ki je pomembno za različna področja, kot so finančni trgi in spremljanje okolja. Za učinkovito izpolnjevanje različnih zahtev smo razvili več vsestranskih algoritmov, ki na podlagi konteksta in operativnih omejitev v realnem času izberejo najprimernejši model za segmentacijo slik, medtem ko brezpilotno letalo leti nad kmetijskim poljem. Naši algoritmi dinamično izbirajo med več obrezanimi (angl. pruned) različicami nevronske mreže U-Net - od 25% do 100% zmogljivosti prvotnega modela - in uravnotežijo natančnost glede na porabo virov. Naš pristop se je izkazal za učinkovitega, saj dosega ali presega metode s fiksnimi modeli. Na primer, naš bandit z zgornjo mejo zaupanja (upper confidence bandit) dosega enako metriko preseka nad unijo (IoU) kot 25% obrezan model, vendar s 15% manjšo uporabljeno težo, kar zmanjša porabo energije. Naši algoritmi se lahko prilagodijo različnim operativnim potrebam kot sta podaljšanje časa letenja drona ali maksimiziranje natančnosti segmentacije, odvisno od uporabnikovih zahtev.

Keywords:spodbujevalno učenje, brezpilotno zračno letalo, napredno kmetovanje, računalniški vid

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back