Process for the classification of objects

Details Process for the classification of objects Process for the classification of objects

1995-10-10

1999-09-07

Hafiz, Tariq R.

Data processing: artificial intelligence

Adaptive system

706 16, 706 17, 706 25, G06F 1518

Patent

active

059501804

DESCRIPTION:

BRIEF SUMMARY
BACKGROUND OF THE INVENTION

The present invention relates to a process for the classification of objects using an electric signal receiver which scans the objects to be classified and delivers for each object M scanned values, which are applied to an evaluation unit.
At many stages of an industrial production process, previous identification is required for adequate processing or inspection of a workpiece. The employed identification process should have a high as possible identification rate while at the same time a low as possible number of false classifications. The parts to be identified have to be identified independent of their position or their orientation. Moreover, soiling or deviation from the desired shape must not significantly influence identification performance. Furthermore, the system has to be able to compensate for slow changes in the ambient conditions, such as, e.g. light, due to a followup learning process during identification.
For economical reasons, the hardware of the identification system has to be as simple as possible; nonetheless sufficiently short identification times have to be feasible.
Another significant problem in automatic production is classification; classification algorithms are an important tool in quality control: good and poor pieces of a production process have to be separated. Usually, in particular, the error rate in failing to recognize poor pieces (if need be at the expense of the recognition rate of good pieces) has to be kept small.
Classification algorithms have to have the same insensitivities as the identification algorithms, by way of illustration, regarding position and orientation of the workpiece, their illumination and soiling, etc.
Hitherto either conventional pattern processing procedures or neuronal nets have been utilized for identification or classification.
For acceleration, conventional processes and neuronal nets usually employ features extracted from the pattern data. Then a classification or identification algorithm continues to work with the aid of these extracted features. It is disadvantageous is first of all, that the selection of problem-specific features is relatively complicated; and secondly, the followup learning capability during the process is small.
Neuronal nets have been used in an attempt to solve the problem of the lack of followup learning. However, with neuronal nets (as with learning input), learning times have to be taken into account which, with the presently available hardware, prevent use in a real-time identification system.
Conventional identification and classification process, therefore, require either complicated hardware for the image recording system and the evaluation system or show relatively poor identification respectively classification performance.


SUMMARY OF THE INVENTION

An object of the present invention is to provide a process for classifying objects using an electrical signal receiver, which scans the object to be classified and delivers for each object M scanned values, which are applied to an evaluation unit, which permits a pattern recognition which meet industrial needs.
A solution to this object is provided by the present invention, as follows.
Further embodiments of the present invention are also provided.
In the process of the present invention for the classification of objects using an electric signal receiver which scans object to be classified and delivers for each object M scanned values, which are applied to an evaluation unit, has a learning step in which by minimization of a potential function for each of i object classes an adjoint prototype is learned, in which class-specific features are contained intensified and suppressed for differenciating irrelevant between i object classes, and a classification step, in which with the aid of another potential function an object to be classified is unequivocally assigned to one of the learned adjoint prototypes.
An element of the present invention is that in order to learn the adjoint prototype per class, j>1 learning objects are utilized

REFERENCES:
patent: 4965725 (1990-10-01), Rutenberg
patent: 5216750 (1993-06-01), Smith
patent: 5295197 (1994-03-01), Takenaga et al.
patent: 5555439 (1996-09-01), Higashino et al.
patent: 5619589 (1997-04-01), Otsu et al.
Hermann Haken, "Synergetic Computers and Cognition, A Top-Down Approach to eural Nets", pp. 36-59, (1991).

Affiliated with

Also associated with

Rating

Process for the classification of objects does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Process for the classification of objects, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Process for the classification of objects will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-1815352