Sensor Technologies - Artificial Human Sensors: Science and Applications - page 99

models, frequently used in artificial sensor system examples, ANN and Fuzzy
Based Techniques, Loutfi (2006), have frequently been illustrated in classifica-
tion applications. Also the emerging interest for using wavelet transformation,
Thuillard (2001), in data analysis is gaining an increasing interest in artificial
sensing applications, e.g., in food quality analysis, Robertsson (2005).
— Validation is the process to test if an appropriate level of complexity has
been attained and the requested generalised capability of the model is suffi-
ciently dynamic. Basically, the validation process has the intent to confirm by
experiments or tests, i.e., evaluation techniques, the precise extent to which a
particular analysing method has the inquired character. An important aspect
is to find a model whose complexity matches the structure of the measurement
task. In a basic set, validation is typically performed by building a model by
using a set of training data and then estimating the behaviour of that model on
a second set of data, the validation data. The validation data then consists of
samples, measured independently of the training data, however, with the same
equipment and external conditions, Fisher (2007).
— System Intelligence or maybe more correctly, a system's adaptability capabil-
ities (self-adapting models) may be of interest to adopt the principles that
have been earned by learning from a set of observations. It is not only pos-
sible to generalise and apply the gained knowledge but also to correlate, pre-
dict and autonomously correct other, previously unknown, observations. The
fundamental assumption made by using such principles is that the proper-
ties of a sensor may experience changes, that are induced by long or short
term drift, reproducibility or unknown external circumstances. Then, the
basic importance of assuming a system's stability can be maintained by an
“intelligent” adaptation of unforeseen degradation in performance, by for
instance miss-interpretations of the data. The degree of autonomous behaviour
that cope with uncertainty may be an expression of intelligence in a multi-
sensor system capability.
The proposed road map for a multi-dimensional sensor system describes a proce-
dure increasing the security and stability in an overall system performance. The
different analysing stages, as described in the suggested structural road map, may
with advantage be used as separate processes or in other combinations than those
proposed. The main purpose is to always recognise the ability of the system per-
formance, i.e., to find variables, whose trends can be explained and from that expe-
rience, gain knowledge of the system peculiarities and the dynamical effects of the
external environment. By knowing the system qualities, then there will be a natu-
ral possibility to logical explanation and understanding of erroneous occurrences
that may happen.
As stated earlier, this topic has in no way the intention, to explain the dif-
ferent possible procedures when using pattern recognition, but rather direct in-
terested readers to a number of illustrative references. General concepts related
to identification and learning from data is given in Theodoridis (2006), Kroenke