corkinspect
success story
GOTTA
LOTTA BETTER STOPPERS
An automatic vision system for the
inspection and grading of cork stoppers.
Barcelona,
3-2003
Summary
Due to the difficulty of inspecting cork, coupled
with high production rates, even the most experienced
inspectors frequently make mistakes. The need for
the automatic inspection of cork is clear. CorkInspect
is a system for the automatic inspection and grading
of cork stoppers.
The current system classifies four stoppers per second
using a linear camera to inspect the lateral area
of the stopper and two matrix cameras for the bases.
This system enables both the real-time classification
of stoppers into several user-defined grades and the
automatic detection of critical defects.
CorkInspect minimises subjectivity in the inspection
of cork stoppers, it improves quality and it reduces
waste through the optimal use of raw material. CorkInspect
also reduces inspection times by up to 60%. By improving
productivity, a system pays for itself in 25 months.
The benefits
The CorkInspect system brings numerous competitive
advantages to the cork industry. It minimises subjectivity
in the inspection of cork stoppers, it improves quality
and it reduces waste through the optimal use of raw
material. CorkInspect also reduces inspection times
by up to 60%. By improving productivity, a system
pays for itself in 25 months.
Background
Inspection is currently the least automated task in
the production of cork stoppers, every one of which
must be inspected. Due to the difficulty of inspecting
cork, coupled with high production rates, even the
most experienced inspectors frequently make mistakes.
Furthermore, it is becoming increasingly difficult
to find workers willing and able to do a job that
is both skilled and highly repetitive. Conversely,
human inspection leads to a lack of objectivity and
uniformity. The need for the automatic inspection
of cork is clear. However this is a challenging application
due to the high production rates, the nature of the
product and the large range of cork features that
are used for classification.
The problem
Although humans can easily detect different defects
in cork and classify stoppers into different grades,
it is difficult to formulate precise rules due to
the variability of the raw material. Even experts
in cork grading find it difficult to define those
features that they use for classification. Because
of this and because a human inspector's performance
can vary widely as a result of visual fatigue and
other distractions, the manual grading of cork stoppers
has a 30% error rate.
It is unacceptable to have an expensive wine bottle
with a defective stopper. Because of this, the accepted
practice in classification is to err on the side of
downgrading good material, rather than upgrading poor
material. Incorrect classification consequently results
in a waste of material.
The problem to be addressed is therefore to find
objective classification parameters and decision rules
in order to implement a vision-based automatic inspection
system. By overcoming the shortcoming of human operators,
such a system can increase productivity and profitability
in the manufacture of cork stoppers.
The solution
To address subjectivity in the process of cork inspection
and grading, different techniques from the fields
of computer vision and pattern recognition have been
used. Machine learning algorithms have been applied
to analyse a set of 43 different features in cork
such as the number of holes and average properties
of different grades of stopper. These algorithms allow
automatic programming of the inspection system, which
learns "by example" to classify stoppers.
The current system classifies four stoppers per second
using a linear camera to inspect the lateral area
of the stopper and two matrix cameras for the bases.
The camera is connected to a computer which classifies
the stoppers automatically. This system enables both
the real-time classification of stoppers into several
user-defined grades and the automatic detection of
critical defects.
Contact details
Petia Radeva
Centre de Visió per Computador
Edifici O, Campus Universitat Autònoma de Barcelona
08193 Bellaterra (Barcelona) Spain
Tel: +34 93 581 18 62
Fax: +34 93 581 16 70
email: petia@cvc.uab.es
