WO2012147017A1

WO2012147017A1 - Method for classifying articles, in particular buttons and/or similar washers, automatic selection process for articles comprising such a method and machine for automatic selection of articles actuatable on the basis of such a process

Info

Publication number: WO2012147017A1
Application number: PCT/IB2012/052001
Authority: WO
Inventors: Michele CAVALOTTI
Original assignee: Tullio Giusi S.P.A.
Priority date: 2011-04-26
Filing date: 2012-04-20
Publication date: 2012-11-01
Also published as: ITMI20110696A1

Abstract

A method (MC) for classifying buttons (A) comprises an automatic learning phase (B) for the classification of a sample population of buttons (A). The learning (B) comprises: acquiring (AI) an image (I) of each button (A) and defining an area (AR) in each image (I); making a first colorimetric quantification (Ql) of each image (I) and identifying a respective first plurality of colours (CI); determining a first representation (Rl) of each button (A) on the base of the first quantification (Ql); making a second colorimetric quantification (Q2) of a summary representation (Rr) by identifying a second plurality of colours (C2); determining a second representation (R2) of every button (A) on the base of the second quantification (Q2); grouping (C) the buttons (A) on the base of the second representation (R2) to obtain a plurality of models representing (Mr) the sample population of buttons (A).

Description

METHOD FOR CLASSIFYING ARTICLES, IN PARTICULAR BUTTONS AND/OR SIMILAR WASHERS, AUTOMATIC SELECTION PROCESS OF THE ARTICLES COMPRISING SUCH A METHOD AND MACHINE FOR THE AUTOMATIC SELECTION OF THE ARTICLES ACTUATABLE ON THE BASIS OF SUCH A PROCESS

The present invention refers to a method for the automatic management of the classification of articles, in particular buttons and/or similar washers.

The present invention also concerns an automatic selection process of articles, in particular buttons and/or the like, comprising the aforementioned method for the automatic management of the classification of the articles.

A further object of the present invention is a machine for automatically selecting articles, in particular buttons and/or similar washers, which can be actuated on the base of the aforementioned automatic selection process and/or the method of the automatic management of the classification of articles.

The object of the present invention is suitable for being used in the industrial field of machines for manufacturing and machining buttons and, in particular, it concerns digital systems associated with the machines for selecting washers, buttons and/or similar articles .

As it is known, the selection of washers, buttons and/or similar articles is currently carried out by means of appropriate machines provided with corresponding digital systems for analysing and identifying the external appearance of the articles being selected.

The selection of the aforementioned articles can be set based on different diversification criteria, like for example, the uniformity of the colour in the passing spot, which can be carried out normally on buttons and/or washers made from horn or from urea, the variation of the shade, which can generally be carried out on buttons and/or washers made from corozo.

The selection of the articles is normally actuated through a machine having a supply station at which the articles to analyse are charged. The machine comprises one inspection plane on which optical means for detecting the images able to detect the external appearance of the articles in exam are operatively active. A conveyor belt projects from the inspection plane, said belt being suitable for transferring the analysed articles along a predetermined forward movement direction. Along the conveyor belt a plurality of transverse conveyors is arranged each of which ends at a substantially box-shaped storing magazine. For each transverse conveyor it is moreover provided, at the conveyor belt, a diverting member which, when suitably activated, deflects the advancing article moving towards the corresponding transverse conveyor and, consequently, towards the respective storing magazine. At the opposite side of the inspection plane, the conveyor belt ends at a further substantially box- shaped storing magazine.

In order to ensure the correct management and programming of the mechanical, pneumatic and electric/electronic members provided and necessary for the movement and the actuation of all the mobile and operative components, the machine is provided with a programmable electronic control unit, like for example a PLC or a similar electronic processor.

In order to allow the actuation of a selection process of the articles during their supply and a subdivision thereof on the base of one or more predetermined criteria, the machine is provided with an electronic selection system that is operatively associated with the electronic control unit.

It is necessary to actuate a preliminary phase of adjusting the exposure and focusing of the optical detection means of the images, so that the electronic selection system can operate to its full potential.

The adjustment of the exposure and of the focusing is obtained by positioning an article in a specific detection area identified on the inspection plane.

Subsequently, at least one lighting device that is operatively associated with the optical detection means is activated so as to improve the quality of the detection to be carried out.

The detection area is provided with a suitable centring mechanism which allows the positioning of the articles on the inspection plane with respect to a reference axis of the optical detection means.

The optical detection means are moreover able to move closer to and further away from the inspection plane so as to enlarge and reduce the image of the article in exam, visualised on a respective reference display.

Subsequently to the centring of the respective article in exam, the adjustment of the exposure and of the focusing of the optical detection means of the images is actuated. The adjustment is generally carried out manually on the respective optics until the image of the reproduced article is clear and well defined on the reference display.

Once the exposure and the focusing of the optical detection means have been adjusted there is the recording of the image .

Once the recording has been carried out a learning phase, necessary for the setting of the selection criteria of the machine, is actuated.

The learning is actuated on the base of one or more significant parameters which can be detected through the optical detection means .

In particular, the learning is carried out on the uniformity of the extraction of the colour relative to the detected image of one or more articles in exam.

Initially, a specific research area is defined and selected on the image detected and is recorded by the aforementioned optical means for detecting the images . The selection of the research area is carried out by means of the management of the width of a circular area which can be enlarged or reduced according to the requirements. Usually, the research area is sized in a way such as to circumscribe the article in exam without passing beyond the outer edge of this latter. Once the research area has been defined, there is then the storing of one or more significant values and/or parameters detected in such an area.

In order to allow the machine to select the articles according to the dimensions of irregular areas, called "spots", it is necessary to preliminarily give each comparing model at least one range of values, relative to a corresponding type of "spot" .

Once a first type of significant parameter of the "spots" has been selected, it is necessary to set a maximum pixel value to give to a first comparing model of a plurality of models to be set.

Subsequently, a second type of significant parameter of the "spots" corresponding to a minimum value of pixels to be given to the first comparing model, is selected. Once such values have been given, there is then the defining of the second comparing model of the aforementioned plurality of models to be set.

The definition of the second comparing model is carried out by initially copying the first comparing model and modifying it at a later moment.

The modification of the copied comparing model is actuated by acting upon the adjustment of the research area the image of which is subsequently recorded.

On the base of such an image both the maximum pixel value and the minimum pixel value to be given to the second comparing model are then set, similarly to what is carried out for the first comparing model.

In detail, the assignment of the maximum and minimum pixel values to the second comparing model is carried out by initially selecting a first type of parameter followed by the setting of the maximum pixel value and, subsequently, by selecting a second type of parameter followed by the setting of the minimum pixel value.

Once the aforementioned values have been given also to the second model, the same is carried out for a series of further comparing models necessary for the correct selection of the articles.

The group of all the set comparing models is stored thus ending the programming of the comparing parameters .

With particular reference to the selection of horn buttons, which have, in some cases, one or more white spots, it is possible to preselect the latter before moving on to the analysis of the dark spots .

With the first selection, by using for example the first and the second comparing model, it is possible to verify the presence of a white spot and define its width.

Subsequently, by using, for example, a further comparing model, it is possible to analyse the width of a dark spot .

In detail, such a process is carried out by placing a button, in which there is a white spot, on the inspection plane.

The exposure and the focusing of the optical detection means are adjusted and the image is subsequently recorded as described above .

There is then the definition of the first comparing model through an operation of extracting the colour from the white spot.

Subsequently, the research area and the width of the white spot are defined.

There is then the definition of the second comparing model on the base of the modification of the data copied from the first comparing model. In such a way the comparing models dedicated to the white spot are set .

Subsequently, it is necessary to vary the colour to be analysed so as to make the selection in relation to the dark spot .

In detail, it starts out from an image of the button that has been previously recorded, on which it is possible to select the dark spot of interest.

The same is carried out, by extracting the colour of the dark spot .

Subsequently, the research area and the width of the spot are defined.

Afterwards, the following comparing model is set by using the comparing model that has just been set and by repeating the extraction cycle of the colour and defining the width of the spot for each following comparing model to be set .

Although the process and the system described above make it possible to obtain a satisfactory selection of the buttons, washers and/or similar articles, the Applicant has found that these still have some drawbacks and can be improved in many aspects, mainly related to the programming speed and simplicity, the streamlining of the learning phase, as well as the accuracy during the automatic selection of the articles .

In particular, the Applicant has found that in the known process and system it is necessary, for every comparing model, to identify the reference pixels to be counted and to set the maximum and minimum thresholds, that is to say the maximum and minimum number of sample pixels used for assigning a general sample. The identifying and assigning phases of the maximum and minimum thresholds inevitably requires the succession of a series of phases that must be carried out step by step by the worker in charge, substantially lengthening the overall time needed for setting the parameters necessary for the correct operation of the program selection.

It is also be considered that the setting of the comparing parameters depends exclusively upon the settings that the worker gives each comparing model on the base of the visualization of the articles on the control display. Therefore, it is not to be excluded that the parameters inserted directly by the worker do not perfectly match the external appearance of the articles in exam, but are close, although in a precise manner, to the real appearance thereof.

In such a way, the automatic selection of the articles, based on such comparing models, has a level of accuracy that is not completely satisfactory since it does not correspond to the real external appearance of the existing articles.

The main purpose of the present invention is to provide a method for the automatic management of the classification of articles, in particular buttons and/or similar washers, an automatic selection process of the articles comprising such a method for the automatic management of the classification of the articles and a machine for the automatic selection of the articles that can be actuated on the base of such a process, that are capable of solving the problems found in the prior art . One purpose of the present invention is to provide a method for classifying articles in a completely automatic manner, without requiring a worker to consecutively set one or more significant parameters for each comparing model to be defined.

A further purpose of the present invention is to provide a selection method that can be carried out on articles during their continuous supply, that is to say without interruptions of the forward flow thereof.

Another purpose of the present invention is to provide a method for classifying the articles that is particularly accurate.

A further purpose of the present invention is to propose a selection process that is accurate and precise .

The purposes specified above, and yet others, are substantially achieved by a method for the automatic management of the classification of articles, in particular buttons and/or similar washers, an automatic selection process of the articles comprising such a method for the automatic management of the classification of the articles and a machine for the automatic selection of articles that can be actuated on the base of such a process, as expressed and described in the following claims.

We shall now describe, as an example, a preferred but not exclusive embodiment of a method for the automatic management of the classification of articles, in particular buttons and/or similar washers, an automatic selection process of the articles comprising such a method for the automatic management of the classification of the articles and a machine for the automatic selection of articles that can be actuated on the base of such a process, in accordance with the present invention.

Such a description shall be made in the rest of the description with reference to the attached drawings, provided only as an indication and therefore not with limiting purposes, in which:

figure 1 is a perspective view of a machine for the selection of articles, in particular buttons and/or washers ;

figure 2 is an enlarged perspective view of a detail of the machine according to figure 1;

figure 3 is an enlarged view of a detail of the machine according to figure 1;

figure 4 is a schematic view of an operative display- that can be used along with the machine according to figure 1;

figure 5 is a block chart relating to the actuation of the machine according to figure 1, in which it is first necessary to carry out a classification method of a sample population of articles so as to actuate a respective process for selecting the articles;

figure 6 is a block chart of the steps of the classification method according to figure 5 ;

figure 7 is a schematic view of a graphic strip corresponding to a first representation of the method according to figures 5 and 6, of a respective sample article ;

figure 8 is a schematic view of a summary representation of a sample population of articles. With reference to figures 1 to 3 , reference 1 wholly indicates a machine for selecting articles A, in particular buttons and/or washers, in accordance with the present invention.

The machine 1 comprises a supply station 2 at which the articles A to be analysed and selected are charged.

Near the supply station 2, the machine 1 is provided with an inspection plane 3 (figures 1 to 3) , at which optical detection means 4 of the images (figures 1 and 3) , optionally a tactile camera 4a are operatively disposed.

Preferably, as illustrated in figure 4, the optical detection means 4 of the images are suitable for detecting the external appearance of the articles A in exam the image I of which can be reproduced on an operative display D to be electronically and digitally treated .

Machine 1 also provides main transfer means 5, in particular a conveyor belt 5a, operatively disposed downwards of the inspection plane 3 for moving away the articles A coming from the latter.

Along the main transfer means 5, at least two collecting conveyors 6, optionally a plurality of collecting conveyors 6, are furthermore distributed each of which has the purpose of leading one or more articles A towards at least one respective substantially box-shaped and/or tray-shaped collecting magazine 7.

As shown in figure 1, each collecting magazine 7 is arranged at a respective collecting conveyor 6 on the opposite side with respect to the main transfer means In order to direct one or more advancing articles A along the main transfer means 5 towards at least one respective collecting conveyor 6 and, consequently, towards a respective collecting magazine 7, the machine 1 is equipped with diverting means 8, optionally a plurality of bellows and/or blowing nozzles 8a, that are connected to a compressed air supply network.

Each bellow and/or blowing nozzle 8a is arranged along one side of the main transfer means 5 according to a position that is aligned with a respective collecting conveyor 6 so as to push at least one article A, by means of an air jet, forward along the main transfer means 5.

In detail, the bellows and/or the blowing nozzles 8a are operatively connected to respective known types of solenoid valves, which are, in turn, connected to a compressed air supply network.

The machine also provides an auxiliary collecting conveyor 9 arranged, on the opposite side with respect to the inspection plane 3, at the terminal portion of the main transfer means 5 so as to lead the articles A, which are not diverted, towards an auxiliary collecting magazine 9 (represented in figure 1) .

Advantageously, the machine 1 comprises at least one programmable electronic unit 10 that is suitable for operating a method C (figure 5) for the classification of the articles A, in particular buttons and/or similar washers, as well as a selection process PS (figure 5) of the articles A charged in the supply station 2.

Advantageously, the programmable electronic unit 10 is charged and provided with a computer program that comprises code portions for implementing the aforementioned method and/or the aforementioned selection process of the articles A. Such a program is advantageously charged in the internal memory of the programmable electronic unit 10.

With particular reference to figure 6, the method MC comprises a learning phase B for the classification of a sample population of articles A.

The learning phase B can be of the automatic or of the assisted type.

The learning phase B of the automatic type can be carried out by images detected through the optical detection means 4 or by images stored in special storage devices, such as for example "Compact Flash" memory cards .

The automatic learning phase B comprises a series of sub steps Impl, AI, D1, Q1, R1, DRr, Q2 , R2, C and P. In particular, the automatic learning phase B comprises a setting phase Impl of at least one classification criterion, in particular representative of the colour, to be used for the classification of the sample articles A.

The learning B further comprises a phase AI of acquiring an image I (figure 4) of each sample article A, which is followed by a definition phase D1 of at least one research area AR (figure 4) in the acquired image I .

Advantageously, the acquisition phase of the image I is carried out by using the aforementioned optical detection means 4 of the images or by pre-charging the images I from a folder or a similar area for storing data directly on a respective memory support, preferably on a "Compact Flash" memory card.

When the optical detection means 4 comprise at least one camera 4a or a similar optical detection means of the images, before moving on to acquiring the image I and to defining the research area AR in the latter, there is a phase of setting the exposure of the image, that is to say the setting, of the diaphragm, of the focusing, of the "shutter" and of the white balance. In order to obtain the best results in terms of definition of the images, there is then a setting of the exposure such that the brighter sample articles A are not overexposed allowing at the same time the distinction of the darker sample articles A belonging to distinct classes.

In this setting phase also the overall number of the sample articles to be analysed for actuating the learning B, which must be representative of the entire population of articles to be selected, can be set.

The learning B also comprises a phase of carrying out a first quantification Q1, in particular a colorimetric quantification, of the acquired image I, through which it is possible to identify a first plurality of colours C1, preferably corresponding to around ten colours that are more common in the acquired image I. Subsequently to the first quantification Q1, a first representation R1 (figure 7) , optionally a graphic representation, of each sample article A is determined on the base of the first quantification Q1 carried out. The first graphic representation Rl contains the centres of the colours of the first plurality of colours CI, as well as the respective percentages of recurrence thereof.

Advantageously, the first representation Rl of each sample article A consists of the representation of a respective graphic strip (figure 7) divided into a plurality of areas S that are aligned along a common direction and that are arranged according to consecutive positions. Preferably, each area S corresponds to a colour of the first plurality of colours CI identified in the acquired image I.

According to one advantageous aspect of the present invention, each area S (figure 7) of the graphic strip of every first representation Rl has a surface that is substantially proportional to the recurrence of a respective colour of the second plurality of colours C2 present on the respective sample article A in exam.

The learning B also comprises a phase DRr of determining a summary representation Rr (figure 8) of the sample articles A by associating and/or grouping the first representations Rl of these latter. Advantageously, the summary representation Rr consists of the graphic representation of a block defined by the group of the graphic strips of each first representation Rl located alongside one another.

The learning B further comprises a phase of carrying out a second quantification Q2 , in particular a colorimetric quantification, of the summary representation Rr, by means of which a second plurality C2 of colours is identified.

More in detail, the second quantification Q2 comprises a phase of associating each colour of the first plurality of colours C1 to at least one corresponding colour of the second plurality of colours C2.

Subsequently, the second quantification Q2 also provides a phase of calculating the recurrence, preferably expressed in a percentage value, of each colour of the second plurality of colours C2 that can be detected in each first representation Rl and a phase of associating, to each sample article A, at least one vector, the components of which correspond to the recurrences of the colours of the second plurality of colours C2, previously calculated.

In particular, according to an example of application of the sectoring of each sample article A, the following table is then compiled:

which lists, in the first row from left to right, the colours of the second plurality of colours C2 , that is to say the colours of the population of sample articles A and in the first column, from top to bottom, each sample article A analysed.

The cells defined by the intersection between colours of the population of the sample articles A analysed and every single sample article A analysed express, in a percentage value, the recurrence of the respective colour of the population of articles present in the respective sample article A.

With the table shown above it is possible to construct a normed space, in which each sample article A is expressed as the vector of the relative percentages containing the colours of the population according to the second quantification Q2.

Advantageously, during the learning phase B, it is possible to set both the number of colours C1 of the population of colours, and the total number of colours C2 of the second quantification Q2.

Subsequently, a second representation R2 of each sample article A is determined on the base of the second quantification Q2.

Advantageously, there is then a grouping phase C of the sample articles A, on the base . of the second representation R2 , in order to obtain a plurality of models Mr representing the sample population of articles A.

In detail, group C comprises at least one identification phase of each sample article A in a space of recurrence of the percentages of colours of the second plurality of colours C2, having a predetermined number of dimensions, preferably twelve. Advantageously, the grouping phase C consists in applying at least one grouping and/or clustering algorithm to the sample articles A identified in the aforementioned space of recurrence.

Even more advantageously, the grouping and/or clustering algorithm consists of a clustering algorithm of the agglomerative kind.

In detail, the grouping and/or clustering algorithm is carried out on the base of the mutual distances attributed to each sample article A in the identified and established space.

Advantageously, the application of the grouping and/or clustering algorithm comprises at least one phase of setting a maximum reference distance followed by a phase of detecting the mutual distances between the sample articles A identified in the aforementioned space. Subsequently, the grouping and/or clustering algorithm identifies the minimum distances between the sample articles A. After that, there is a phase of grouping the sample articles A relative to the minimum distances detected. Subsequently, the average point of the minimum distances and the attribution, to the latter, of a double weight with respect to the weight of the corresponding sample articles A, is identified. At this stage it is possible to eliminate the sample articles A which, on the base of a relative comparison between the mutual distances and the maximum reference distance, are particularly far from the reference parameters .

It is then possible to move on to calculating the centres and the radiuses of the new classes identified through the agglomerative and/or clustering algorithm so as to publish and communicate P the significant data obtained to the worker.

According to an example of a preferred application of the agglomerative and/or clustering algorithm the mutual distances attributed to each article A identified in the aforementioned space are inserted in the following matrix:

where both the first row from top to bottom and the first column from left to right have the list of sample articles A analysed, whereas the central cells, corresponding to the intersections between the rows and the columns, express the relative distances of the sample articles A identified in the aforementioned normed space .

Advantageously, in accordance with such an example of application of the agglomerative and/or clustering algorithm, the mutual distances of the sample articles A in the space of the aforementioned percentages of colours, can be expressed according to the following equation:

The learning B of the assisted type is identical to the automatic type, from the sub step of setting impl of at least one classification criterion representative of at least one colour to the sub step of determining a second representation R2 of each sample article A on the base of the second quantification Q2. In the assisted learning the grouping C of the sample articles A is carried out based upon the organization of the samples A inside special memory addresses, that are preset in the relative memory supports used, like for example the so called "Compact Flash" memory cards or the like.

Advantageously, during the aforementioned learning phase B it is also possible to set:

- the number of selection classes;

- the minimum percentage for creating a selection class ;

- the learning modality B to be carried out between automatic or assisted;

- the saving parameters of the learning images I, as well as the saving path thereof.

In accordance with a preferred solution of the present invention, the agglomerative and/or clustering algorithm, also called selection algorithm is strictly correlated to different parameters and/or actuation modalities. In particular, the aforementioned algorithm varies according to the automatic weights, to the automatic radiuses and to the type of learning (automatic or assisted) .

With reference to the automatic weights, the Applicant has noted that small variations in the amount of colour of the population of colours, like for example situations in which the colour is scarcely present but very far from most of the remaining colours, these are not taken into account by the software. It is thus necessary to represent a range of twelve real colours of the population in exam, through the conversion from LAB to RGB, and it is necessary to assign to each colour a respective weight.

After determining the aforementioned colours it is then possible to move on as follows:

1) automatic quadratic modality:

2) automatic linear modality:

3) unitary modality:

4) manual modality, in which it is possible to adjust the weights individually:

After determining the eight centres of the eight classes, it is then possible to move on as follows:

1) i-th centre coordinates

2) i-th standard deviation

in which ni is the number of samples of the i-th class. In attribution the distance dz is calculated between every sample z and the centre of the i-th class and the nearest class s (with lower d) is determined. In the case in which ds<rs it is attributed to the class s, otherwise it is discarded, as follows:

In this case the radius of the i-th class, without automatic radiuses, is:

With reference with the automatic radiuses, at the end of the learning B the radiuses of the classes are determined .

In the case in which the automatic radiuses modality is enabled, such radiuses start from a lower value:

For each i-th class the nearest class "v" is determined, as follows:

in which n represents the number of classes.

As an example, if we consider class 1 and we hypothesise that from the calculations obtained we have that the closest class is class 2, then according to the relation i=l; and v=2, the distance between class 1 and 2 would correspond to:

In the case in which r₁+r₂>D₁₂, that is to say if

the classes permeate each other. It is thus necessary to suitably scale them.

The redefining of the radiuses must take into account the standard deviation of the considered classes (ΟΊ and σ₂) and the number of samples of the considered classes (N₁ and N₂) .

The class which has the lowest number of samples requires a greater reduction with respect to the class that has the highest number of samples.

Therefore, the class with lowest σ must be reduced more with respect to the class with greater σ.

In order to do this, it is necessary to introduce suitable respective weights.

In addition, it is preferable for K₁ · o₁+K₂ · o₂=D₁₂ , where K₁ and K₂ are respective coefficients which must consider the number of samples, according to the following relationships:

where α represents the coefficient of overall reduction.

Advantageously, the coefficient of overall reduction α is determined in the following manner:

consequently the new constants are

Therefore the new resized radiuses are:

For each of the eight classes it is calculated which of the other seven centres is the centre of the closest class. Consequently, for each of the provided eight classes a partition of the relative distance is established as radius or rather a weighted average from the minimum distance is established as the radius based on the number of samples and on the standard deviation. Through the automatic resizing of the radiuses, the singularity of belonging to the class is ensured, that is to say, a button can belong to one and only one class. Therefore, a button is always attributed to the same class.

In the case in which the automatic resizing of the radiuses is disabled, all the radiuses take up the value 2a, so that for 95% of the samples of the i-th class is comprised in the range:

A further object of the present invention is a selection process of the aforementioned articles A which comprises a learning phase B which provides the actuation of the aforementioned method A of automatic classification of the sample articles A.

Once the automatic classification of the sample articles A has been carried out, the process provides the actuation of the selection which comprises, for each article being supplied, a phase of inspection of the external appearance . There is then a phase of subdividing the articles A being supplied into at least two groups, preferably a plurality of groups, based upon the selection criteria and upon the classes of data set through the automatic classification method.

Of course, as mentioned above, it is possible to establish and set the source to be used which can therefore correspond to the camera or to any memory support, preferably a "folder", on which the required data and/or the images to be used are stored.

In the case in which the aforementioned memory support is used, it is possible to also carry out a selection simulation without actually making the latter.

The subdivision of the articles A is followed by a grouping phase of the similar articles according to one or more criteria or classes of common data previously established.

In accordance with one example of operation process, the system tries to attribute, to one of the classes found during the learning, each of the examples which are proposed to it by using the colorimetric distance from the centres of the classes and the radiuses of the latter .

Once the first eight classes have been separated it is possible to distinguish what remains and the discards by examining a representative number of samples of this new population in a new learning process.

If the algorithm groups two distinct categories of articles A in a single class, the worker can subdivide them by examining a representative number of samples of this new population in a new learning process. Alternatively, it is possible to store the images directly on the data processor or PC, or on any memory support, by organizing them in distinct folders, each relative to a different class, or again pre-charging the images directly on "Compact Flash" memory cards and launching an assisted learning.

The method, the process and the machine described above solve the problems found in the prior art and achieve important advantages.

Firstly, the level of accuracy in both the learning and classification phases of the sample articles A, which in the phase of selecting the articles is particularly more accurate since the system automatically and precisely sets the parameters relative to the detected colours without requiring the intervention and the setting of any significant parameter by the worker.

The worker, in this case, must only select the research area, set the shutter and the diaphragm and set the number of samples on which to carry out the learning. The worker can improve the results of the automatic learning by optimizing it with a new assisted learning. It should also be considered that according to the object of the present invention it is possible to operate according to two operation modalities. Reading can be carried out with the stopping of the articles or continuous reading can be carried out without stopping of the articles.

In the modality of selecting from a memory support or "folder" it is advantageously possible to obtain a preview of the selection which shall be carried out before actually actuating it.

Claims

1. A method (MC) for classifying articles (A), in particular buttons and/or similar washers, comprising an automatic learning phase (B) for the classification of a sample population of articles (A) , said automatic learning phase (B) comprising following sub steps:

- setting (Impl) at least one classification criterion representative of at least one color, to be used for the classification of said sample articles (A) ;

- acquiring (AI) an image (I) of each sample article (A) and defining at least one research area (AR) in said acquired image (I) ;

- making a first colorimetric quantification (Q1) of said acquired image (I) by identifying a first plurality of colors (C1) and determining a first representation (R1) of each sample article (A) on the base of said quantification (Q1) ;

- determining (DRr) a summary representation (Rr) of said sample article (A) by associating the first representations (R1) of these latter;

- making a second colorimetric quantification (Q2) of said summary representation (Rr) by identifying a second plurality of colors (C2) and determining a second representation (R2) of each sample article (A) on the base of said second representation (Q2) ;

- grouping (C) said sample articles (A) on the base of said second representation (R2) in order to obtain a plurality of models (Mr) representing said sample population of articles (A) .

2. The method according to claim 1, in which said first representation (Rl) of each sample article is a graphic strip divided in areas (S) , each area (S) corresponding to a respective color of said first plurality of colors (CI) and having an extension proportional to the recurrence of this latter.

3. The method according to claim 2, in which said summary representation (Rr) consists in a block defined by the grouping of said graphic strips of said first representations (Rl) .

4. The method according to claim 3 , in which said second quantification (Q2) comprises the phase of associating to each color of said first plurality of colors (C1) a corresponding color of said second plurality of colors (C2) .

5. The method according to claim 4, in which said second quantification (Q2) further comprises the phases of:

calculating the recurrence of each color of said second plurality of colors (C1) in each first graphic representation; and,

associating to each article (A) a vector the components of which correspond to the previously calculated recurrences .

6. The method according to claim 5, in which said grouping phase (C) further comprises an identification phase of each sample article in a space of recurrence of the percentages of colors of said second plurality of colors (C2) on the base of the vectorization of each sample article (A) , said space of recurrence having a predetermined number of dimensions, in particular twelve .

7. The method according to claim 6, in which the grouping phase (C) further comprises the application of at least one grouping and/or clustering algorithm on the sample articles (A) identified in said space of recurrence .

8. The method according to claim 7, in which the grouping and/or clustering algorithm consists in a clustering algorithm of the agglomerative kind.

9. The method according to claim 7 or 8, in which the application of the grouping and/or clustering algorithm is made on the base of mutual distances attributed to each sample article (A) .

10. The method according to any of claims 7 to 9, in which the application of the grouping and/or clustering algorithm comprises the phases of:

setting a maximum reference distance;

detecting the mutual distances among the sample articles (A) ;

identifying the minimum distances among said sample articles (A) ;

grouping the sample articles (A) in relation to the minimum distances detected;

identifying the average point of said distances by attributing to it a double weight with respect to the corresponding sample articles (A) ;

eliminating the sample articles (A) on the base of a relative comparison between the mutual distances and the maximum reference distance;

calculating the centre and the radiuses of the identified classes;

publish (P) the result of the calculation.

11. A computer program directly chargeable in the internal memory of a digital computer (10) , comprising code portions for implementing the method according to any of preceding claims when said program is actuated by said computer.

12. A selection process (PS) of articles (A), in particular buttons and/or similar washers, comprising the phases of:

- actuating a learning process, according to the method (MC) for the automatic management of the classification of the articles (A) according to one or more preceding claims ;

- inspecting a plurality of articles (A) during their supply;

- subdividing said supplied articles (A) in at least two groups, preferably a plurality of groups, on the base of the selection criteria and of the classes of data set with the classification method (MC) ;

- grouping the articles (A) according to one or more criteria or classes of common data.

13. The computer program directly chargeable in the internal memory of a digital computer (10) , comprising code portions for implementing the process according to claim 11 when said program is actuated by said computer (10) .

14. A machine (1) for the selection of articles (A), in particular buttons and/or washers, comprising:

- one supply station (2) at which the articles (A) to analyze are charged;

- one inspection plane (3);

- optical detection means (4) of the images (I) , in particular a tactile camera (4a) , able to detect the external appearance of the article (A) in exam, said optical detection means (4) of the images (I) being operatively disposed at the inspection plane (3) ;

- main transfer means (5) , in particular a conveyor belt (5a) , operatively disposed downwards of the inspection plane (3) ;

- at least two collecting conveyors (6) , optionally a plurality of collecting conveyors (6) , distributed along said main transfer means (5) ;

- at least a collecting magazine (7) for each collecting conveyor (6) oppositely disposed with respect to said main transfer means (5) ;

- diverting means (8) operatively associated to said main transfer means (5) , said diverting means (8) being switchable between a first condition, in which they are at rest, and a second condition, in which they divert at least one of said inspected articles (A) towards a respective collecting conveyor (6) ;

characterized in that it further comprises at least one programmable electronic unit (10) able to operate the selection process (PS) according to claim 12.