Please use this identifier to cite or link to this item:
Title: Smart Farming with Wireless Visual Sensor Networks
Contributor(s): Aseel Hasan Edan Al-Nasar (author); Sadgrove, Edmund  (supervisor)orcid ; Fleming, Peter  (supervisor); Falzon, Gregory  (supervisor)orcid ; Paul, David John  (supervisor)orcid 
Conferred Date: 2022-02-03
Copyright Date: 2021-02
Thesis Restriction Date until: 2023-05-03
Handle Link:
Related Research Outputs:

Agriculture has undergone a major technological shift in recent years. The emergence of Smart Farming and Digital Agriculture has driven a data-fuelled productivity revolution. Internet of Things sensors are a central component of Smart Farming which are providing timely information to increase both the yield and quality of agricultural produce. Smart farming increasingly relies on recording and processing a variety of image data collected from wireless visual sensor networks incorporating field robots and drones, smartphones and remote surveillance cameras. The images collected are used for a wide range of farm management tasks including crop yield estimation, livestock monitoring, plant disease diagnosis and weed control. Limited network bandwidth and power constraints are major obstacles to the successful utilisation of visual sensor technology on the Smart Farm. Agricultural production often occurs in remote regions with limited network infrastructure and the sensors are most often required at distant locations on the farm far from network connection points. Remote visual sensors such as drones or solar-powered cameras are also often battery powered with limited endurance. These constraints place major challenges in the development of Smart Farming technology.

Computationally efficient image compression has the potential to greatly address the bandwidth and power constraints of Smart Farming. The fog computing architecture is well suited for this scenario, farm assets can be monitored remotely via a low bandwidth connection with image compression occurring in field on fog nodes prior to upload to the cloud. Contemporary image compression algorithms are too computationally intensive or provide insufficient compression ratios for many Smart Farming applications. Recent advances in neural network-based image compression algorithms provides promising avenues of investigation to achieve rapid, computationally efficient compression algorithms with high compression ratios with minimal model storage. The Extreme Learning Machine Autoencoder (ELM-AE) is particularly promising due to its shallow network architecture which permits high computational efficiency whilst retaining high compression ratios and excellent signal recovery.

The objectives of this thesis are to develop, apply and evaluate novel image compression algorithms for Smart Farm visual sensor network datasets. The thesis provides the following contributions: (i) implementation and assessment of a novel Extreme Learning Machine AutoEncoder (ELM-AE) for Smart Farm image data, (ii) development and evaluation of the Online Sequential Extreme Learning Machine Auto-encoder (OS-ELM-AE) an online image compression algorithm to address the challenges of dynamically changing scenery and (iii) introduction of the Dynamic Update Online Sequential Extreme Learning Machine (DU-OSELM) to address the saturation problem (loss of β weights matrix adaption) in a time-varying environment using two approaches: dynamic forgetting factor and updated selection strategy. These three algorithms were assessed and benchmarked with other candidate compression algorithms over three relevant datasets (landscapes and scenery, weed detection and leaf assessment). This thesis examines compression paradigms that can overcome significant challenges of transmitting digital imagery in Smart Farm environments (e.g., high resolution, limited bandwidth, hardware and power constraints, dynamic environment, real-time transmission) by exploiting the strengths of machine learning and data analysis in the proposed compression model.

The results obtained in this thesis demonstrate that the Extreme Learning Machine Autoencoder compression algorithm and its variants provides a competitive option in terms of compression ratio, processing time (encoding and decoding) and computational complexity. Transmission of the encoded image to the cloud also enhances privacy and security compared to transmitting the entire image. The Extreme Learning Machine Autoencoder approach has been found to be a promising avenue to compress and transmit image data over low bandwidth connections and Smart Farm wireless visual sensor networks.

Publication Type: Thesis Doctoral
Fields of Research (FoR) 2008: 070203 Animal Management
080106 Image Processing
080401 Coding and Information Theory
Socio-Economic Objective (SEO) 2008: 890201 Application Software Packages (excl. Computer Games)
970108 Expanding Knowledge in the Information and Computing Sciences
970110 Expanding Knowledge in Technology
HERDC Category Description: T2 Thesis - Doctorate by Research
Description: Please contact if you require access to this thesis for the purpose of research or study.
Appears in Collections:School of Science and Technology
Thesis Doctoral

Files in This Item:
3 files
File Description SizeFormat 
Show full item record

Page view(s)

checked on Dec 17, 2023


checked on Dec 17, 2023
Google Media

Google ScholarTM


Items in Research UNE are protected by copyright, with all rights reserved, unless otherwise indicated.