OpenAIR @ RGU >
Design and Technology >
Engineering >
Theses (Engineering) >

Please use this identifier to cite or link to this item:
This item has been viewed 39 times in the last year. View Statistics

Files in This Item:

File Description SizeFormat
Reddipogu PhD thesis.pdf4.86 MBAdobe PDFView/Open
Title: A sensory system for robots using evolutionary artificial neural networks.
Authors: Reddipogu, Ann
Supervisors: Maxwell, Grant M.
MacLeod, Christopher
Elder, Stewart
Issue Date: Aug-2006
Publisher: Robert Gordon University
Abstract: The thesis presents the research involved with developing an Intelligent Vision System for an animat that can analyse a visual scene in uncontrolled environments. Inspiration was drawn both from Biological Visual Systems and Artificial Image Recognition Systems. Several Biological Systems including the Insect, Toad and Human Visual Systems were studied alongside popular Pattern Recognition Systems such as fully connected Feedforward Networks, Modular Neural Networks and the Neocognitron. The developed system, called the Distributed Neural Network (DNN) was based on the sensory-motor connections in the common toad, Bufo Bufo. The sparsely connected network architecture has features of modularity enhanced by the presence of lateral inhibitory connections. It was implemented using Evolutionary Artificial Neural Networks (EANN). A novel method called ‘FUSION’ was used to train the DNN, which is an amalgamation of several concepts of learning in Artificial Neural Networks such as Unsupervised Learning, Supervised Learning, Reinforcement Learning, Competitive Learning, Self-organisation and Fuzzy Logic. The DNN has unique feature detecting capabilities. When the DNN was tested using images that comprised of combination of features used in the training set, the DNN was successful in recognising individual features. The combinations of features were never used in the training set. This is a unique feature of the DNN trained using Fusion that cannot be matched by any other popular ANN architecture or training method. The system proved to be robust in dealing with New and Noisy Images. The unique features of the DNN make the network suitable for applications in robotics such as obstacle avoidance and terrain recognition, where the environment is unpredictable. The network can also be used in the field of Medical Imaging, Biometrics (Face and Finger Print Recognition) and Quality Inspection in the Food Processing Industry and applications in other uncontrolled environments.
Appears in Collections:Theses (Engineering)

All items in OpenAIR are protected by copyright, with all rights reserved.


   Disclaimer | Freedom of Information | Privacy Statement |Copyright ©2012 Robert Gordon University, Garthdee House, Garthdee Road, Aberdeen, AB10 7QB, Scotland, UK: a Scottish charity, registration No. SC013781