Biomedical Engineering Reference
In-Depth Information
divide and conquer large scale TSP has the advantages of increased scalability and
parallelism.
An ART-2 network was used to develop a strategy for the navigation of mobile
robots in uncertain indoor environments [
55
]. More exactly, a modified ART-2
network was put forward to identify the surrounding environment correctly for
mobile robots. Path planning is one of the vital tasks in the navigation of autono-
mous mobile robots and may be divided into two categories: global path planning
based on a priori complete information about the environment and local path
planning based on sensor information in uncertain environments where the size,
shape, and location of obstacles are unknown. Local path planning could be called
reactive strategies. The neural networks and fuzzy controllers have proved to
perform well in reactive navigation applications. Computer simulations were
made for design strategies of environment classifiers based on a modified ART-2
network and fuzzy controller. One more example of an ART application is
connected with a mobile vehicle [
56
]. In this case, an ART network is used for
image processing. On the robot's path are several labels, the letters L, R, B, F, and
S, which represent turning left or right, moving backward or forward, and stopping.
ART is used to recognize them and give out the signal to control a mobile vehicle.
Other experiments were conducted involving obstacle avoidance. Eight obstacle
patterns were selected to train the ART network. The potential to add new cate-
gories is very important for this type of task.
In another study [
57
], the ART family of neural networks was used to develop a
speaker recognition system. This system consists of two modules: a wavelet-
based feature extractor and a neural-network-based classifier. Performance of the
system has been evaluated using the gender recognition and speaker recognition
problems. In the gender recognition problem, the highest accuracy was 90.33%;
in the speaker recognition problem, ART-based classifiers have demonstrated
recognition accuracy of 81.4%.
2.6.3 Self-Organizing Feature Map (SOFM) Neural Networks
Teuvo Kohonen published his first articles in the seventies [
29
]. He applied a
specific type of neural network - Self-Organizing Feature Map (SOFM) designs
with the following self-organization training principles. The principal idea is that a
set of processing elements arrange their weights in such a way that they are
distributed in space with a density approximately proportional to the probability
density of the input vector.
This approach found a place in modern investigations and applications. For
example, a Kohonen self-organizing map can be used for unsupervised segmenta-
tion of single-channel magnetic resonance (MR) images [
58
]. The introduction of
advanced medical techniques, such as MRI, has dramatically improved the quality
of the diagnosis and treatment of brain pathologies. The image information in such
systems is complex and has a great number of dimensions. The availability of
