Sun, Jun
(2024)
Developing Hopfield Neural Networks Using Gaussian Distributed Small World Topology For Visual Object Tracking.
PhD thesis, Perpustakaan Hamzah Sendut.
Abstract
Visual object tracking (vot) is considered a challenging research topic in
artificial intelligence. Today, many industries rely on object tracking technologies to
identify errors, monitor environments, and make timely decisions based on tracking
results. Visual object tracking has enabled many innovations, such as autonomous
vehicles, traffic monitoring systems, remote medical diagnostic systems, and more
cutting-edge applications are on the horizon. However, among these notable
achievements, it is worth noting that, unlike these object-tracking techniques, a human
brain is more efficient for object tracking tasks and requires fewer resources. Recent
neuroscience studies have shown that artificial neural networks organized as real
cortical connectivity may perform more efficiently in complex recognition tasks.
Therefore, a novel visual object tracking method based on hopfield neural networks
is proposed in this study. A small-world network is employed as the topology of the
neural network model. However, a biological feature is integrated into the small-world
network model: the exponential decay rule, which may mimic some characteristics of
the structure of the cerebral cortex. In the neural network, each pixel of video frames
is assigned to a neuron at the corresponding position. Pixel strength is characterized as
the state of a neuron. The video frame is memorized after all neurons in the neural
network have been trained to a stable state. A bionic mechanism utilizing the
associative memory property of a bionic hopfield neural network is proposed to track
objects in video frames.
Actions (login required)
 |
View Item |
