fraudulent products with a high sensitivity and minimum training effort. The E-cookbook calls for each expected ingredient in
a recipe step on the corresponding decision-making unit. New
ingredients and corresponding decision-making units can easily
be added to the database. Even an identification of an unknown
ingredient can be approached by sequentially calling on the
OCC units in the database and finding an acceptable match [ 2].
Further, the raw fingerprint will be subject to feature computation and dimensionality reduction to remove redundancy
and irrelevance for the specific task and obtain a lean yet better
discerning decision-making unit. For the rapid and dependable
extension and adaptation to increasing tasks, our DAICOX-system for the automated design of optimized technical
cognition systems will be employed.
OIl REcOgNItION OR gRADINg
The LoX-devices have been used in numerous tasks, such as
detecting freshness and decay in milk and contamination in
wine; recognizing types of fish or meat; and discerning types,
grades, and conditions of various edible oils.
These capabilities are summarized and partially demon-
strated in the corresponding video [ 2]. Using the LoX-devices to
identify and assess the type, grade, and condition of edible oils
can, in many respects, follow the approach used to assess simi-
lar characteristics in the non-edible oils and lubricants used in
engines, gearboxes, and windmills. Fig. 3 shows the NIR range
spectroscopic data for five different vegetable oils that were
examined, based on 15 samples of each.: ( 1) peanut oil, quality
A, ( 2) soybean oil, ( 3) olive oil ( 4) sunflower oil, and ( 5) peanut
oil, quality B. In spite of scatter within each of the oil categories,
the different varieties of oil can be clearly distinguished from
one other, and the two qualities of peanut oil are also discern-
ible from the similarity or feature space plot. Thus, a fraudulent
oil product could potentially be detected by comparing it to the
fingerprint of the oil that is expected to be used in a particular
cooking step. On the other hand, analyzing the composition and
detailed contents of a fraudulent oil is not within the scope of
the current LoX devices.
FIg. 3. plot showing the similarity and discernibility of
five edible oil samples based on the selection of channel
8 and 28 of the visual and NIR range spectroscopic data,
clockwise from 5 o’clock: quality A peanut oil (purple),
soybean oil (red), olive oil (green), sunflower oil (orange),
and quality b peanut oil (yellow).
FIg. 2. block diagram of the E-taster-assistance-system with lox devices and mobile extension