HK1013038B - Method for optically sorting bulk material - Google Patents

Description

The invention relates to a process for optical sorting of bulk materials in accordance with the general concept of claim 1.

A similar procedure is known from US-A-5-085-325.

It is already known that test material is transported on tapes and that the image is taken for testing by a diode line camera or television camera. Signal recording is preferably in flight, e.g. when the test material is transferred from one tape to another.

In modern imaging, color is also captured, and color is used to detect conspicuous areas in the image.

The image of the test item is evaluated step by step with the image scan so that it can be classified immediately after a test piece has passed through the measuring station, allowing the parts to be ejected in flight by flaps or air nozzles.

A disadvantage of the known methods is that the detection rate for colour heterogeneous products is low if the detection of striking points is limited to the detection of colour values not contained in the product because there are many different colour values in the product.

The purpose of the invention is to improve the method of optical sorting of bulk materials so that detectable foreign bodies with a very low error rate are detected in the case of colourless heterogeneous bulk materials.

The purpose of this procedure is to solve this problem, with the characteristics of patent claim 1.

In the case of image capture, the light from each point is filtered by color before the detection elements of a line into the three color components red (R), green (G) and blue (B), which makes it possible to detect striking points (points with color values that are rarely found in the error-free product) by evaluating the color values measured by the line elements (intensities of the respective color components).

First, the entire range of possible colour values in the colour space is divided into several sub-areas, the colour space being stretched by the different colour components measured for each image point. For example, the three colour components red, green and blue form a three-dimensional colour space. Classifiers, i.e. means of evaluating the measured values on the basis of given criteria, allow a classification of the measured colour values, whereby a classifier concentrates on only one sub-area and in this sub-area detects detection areas, i.e. contiguous areas of conspicuous image points, for the colourally heterogeneous product.

If the colour values of a colour homogeneous part are preferred in the selected sub-area, the colour part is detected as a relatively large area of image points of the colour values of the selected sub-area and can be detected by the classifier by evaluating this detection area. On the other hand, in the case of the error-free product, large areas of conspicuous image points are generally only rarely found within such a sub-area and the number of error detections is thus kept low.

In a preferred embodiment, the distribution of the colour values of the sub-sections in which the components of the committee are assumed is learned by showing the components of the committee.

The following drawings give a detailed description of the invention: Figure 1 shows an example of a one-dimensional colour distribution with areas for good test material.Figure 2 shows an example of classification with parallel classifiers in the detection of a panel whose colour values overlap with the colour values of the product.Figure 3 shows an example of correcting a classifier by retrieval.Figure 4 shows an example of colour distortion on subjects when using a camera with image points where the colour values are adjacent.

In a colour sorting machine, the bulk preferably passes in flight by an observation head with a light source and a product signal receiver located near the light source. The reflected light from each image point of the test item is divided by different colour filters of adjacent line elements of a camera line, e.g. a CCD line, of the receiver into the three colours red (R), green (G) and blue (B). The line elements thus measure the brightness of the image points, including colour values, in their respective spectral ranges.

In a pre-learning process, the test material is measured without cutting parts and the frequency distribution of the colour values is determined.

In a retrieval process, the test substance is also measured without the test components and a first step is to determine a range of colour values for good test substance by setting an empirical threshold 2 over the frequency distribution 1 of colour values, the intersections between threshold 2 and the frequency distribution 1 curve being used to determine the limits of the range of colour values of the test substance.

The choice of threshold 2 will also result in the presence of image points on the test material which are considered to be conspicuous, but which, if they are grouped into large areas, would be erroneously classified as a committee. Experience has now shown that such a grouping occurs in turn in preference in certain colour ranges. To measure these colour ranges, a large area detected in the test material which is free of defects is recorded in the learning process and its colour distribution measured. This distribution is introduced as threshold 3 after standardization.

In the case of the measurement of test material with coated parts, the product colour ranges are divided into sub-ranges. In this example, each of the classifiers A, B and C, which operate in parallel, concentrates on only one sub-range. If the colour ranges of the colour homogeneous coating are preferably in the selected sub-range, the coating is detected as a relatively large area and can be detected by evaluating the detection areas. Here again the colour ranges of these large areas are measured and, after standardization, introduced as thresholds.

It is also possible for a fault-free product to have large areas of detection in a range of colours covered by a classifier and thus the fault-free product is classified as a committee. In a further trace-learning process, specifically, these colour values leading to large areas of detection in the fault-free product range are learned and recognized as good test material by changing the thresholds.

After learning, the product is automatically tested.

The test, which may take days, will involve systematic drift-like changes in the product. These changes will result in system performance decreasing over time. To avoid this, the classification system is doubled. One system takes over the test task while the other system measures the current color distribution of the product. The measurement of the current color distribution is monitored by the testing classifier so that no color values of the panel are recorded in this measurement. After a representative number of measurements are recorded, the learning classifier is activated with the newly measured distribution for the test task, while the classifier set to test takes over the learning task.

This adjustment is only possible if a colour point detected, which is considered to be conspicuous, does not always lead to a committee decision. If a colour point detected always leads to a committee decision, the learner classifier would not be able to take on new colour values, since a committee decision would reject the newly learned colour distribution. However, since the system classifies colour points detected as a committee only if they form a larger area, the measured frequency can also be adjusted for detected values. Conversely, the system with this adjustment committee detects colour values belonging to the colour system, which were no longer represented in the colour distribution of the product in a previous measurement and are no longer included in the current measured distribution.

In the case of signal recording, the test object is illuminated, for example, by two lamps in the direction of the line camera, between which the optical axis of the line camera lies.

This requirement cannot be met if the test piece is taken lying on the conveyor belt. Due to contamination and wear, the belt does not have a uniform colour. In addition, shadows form on the conveyor belt, which in general leads to a significant expansion of the colour distribution when measuring the error-free test piece.

In a first embodiment, the background is the colour of the test material, which has the advantage of low contrast between the background and the test material and therefore does not significantly extend the colour distribution of the test material by marginal effects during the transition from the background to the test material.

The disadvantage of contamination is avoided by making the background rotating roll, which immediately ejected the deposits. The shadow of the test substance on the background becomes diffuse and harmless depending on the density of the deposits when the rotating roller is installed at a suitable distance from the test substance. In case of high density of the test substance, excessive darkening of the background is avoided by additional background illumination. Alternatively, the background may be a cylindrical radiator, which radiates in the colour of the test substance and is surrounded by a transparent rotating roll, which ejected the deposits.

In a second embodiment, the background is a dark hole, which has the advantage of allowing the test piece to be segmented from the background without interference by dirt and shading.

The camera is designed to be as large as possible and with low reflectivity walls, and the line camera looks through a slit in the tank, which is adjusted to its width, aperture, focal length and distance from the camera lens.

In the case of ordinary colour cameras, the colour sensors are even locally adjacent to each other, so that the colour sensors see different locations of the measuring object in relation to one point of the image. In relation to Fig. 4, the colour sensors (R, G, B) are arranged horizontally, while the measuring object moves from top to bottom along this horizontal line.In Fig. 4 only the sensor triplet Xn, Yn measures the correct colour of the measuring object. For all other triplets, colour values are measured which contain at least one colour value darker than the corresponding colour value of the test item. For example, the triplet Xn, Yn-1 measures the P rule R = 50, G = 25 and B = 10.by storing the signal levels and comparing the horizontal and vertical neighbouring points.

Key to the Figures The following table shows the figures:

The test results are then compared to the results of the previous test.

The following table shows the figures:

The Commission has decided to grant the aid to the following undertakings:

The following table shows the figures:

Occupancy densityDetection and detection of reject Committee measured value

The following table shows the figures:

The test subject is examined by a test subject.

Claims (10)

1. A process for the optical sorting of bulk material mixed with reject parts, such as agricultural products, drugs, ores, etc. in a colour sorting machine by this bulk material being conveyed over a transport belt and being moved past an observation head with a light source and a product signal receiver arranged in the vicinity of the light source, whereby the reflected light of the image points of the examination material is broken down by various colour filters of detection elements lying next to one another in a line of the receiver into several colour components and the examination material is sorted on the basis of the colour values corresponding to the measured intensities of the respective colour components, characterised in that in each case the colour values of the product are examined in several selected sub-regions of the colour space by, in each sub-region, a classifier ascertaining connected areas of image points with colour values falling into the respective sub-region and carrying out a classification according to preset criteria from the geometry and size of these detected areas.
2. A process according to Claim 1, characterised in that

   - the examination material is surveyed in a prelearning process without reject parts and its colour value frequency distribution is determined dependent on the colour;

   - in a relearning process, in a first step using examination material without reject parts, a colour value region is defined for good examination material by laying a threshold based on experience over the frequency distribution of the colour values, wherein the limits of the colour value region of the examination material result from the intersection points between the threshold and the curve of the frequency distribution;

   - ascertained in the relearning process, using examination material without reject parts, are measured values suspected of containing foreign objects in dependence on colour, which according to the limits of the colour value region of the examination material from the first step of the relearning process would mistakenly be classified as reject, and the size of the local accumulation of these measured values is determined; and

   - in the relearning process, using examination material without reject parts, when a preset magnitude of this local accumulation of measured values suspected of containing foreign objects is exceeded, the threshold value decision of the first step of the relearning process is changed dependent on colour so that for these measured values a decision is made for good examination material.
3. A process according to one of the preceding Claims, characterised in that during examination of the examination material mixed with reject parts, the classifiers operating in parallel only analyse sub-regions of the colour space in which reject parts are suspected.
4. A process according to Claim 3, characterised in that in the sub-regions of the colour space in which reject parts are suspected, the colour value distribution of reject parts is learnt by displaying them.
5. A process according to one of the preceding Claims, characterised in that the examination of the examination material mixed with reject parts takes place with a first classification system, while the current frequency distribution of the colour values of the examination material is measured for adaptation to systematic drift-like changes of the examination material with a second classification system, the measurement by the examining first classifier being monitored so that no reject values are detected during the measurement.
6. A process according to Claim 5, characterised in that both classification systems alternate in their function.
7. A process according to one of the preceding Claims, characterised in that by comparing colour signals of adjacent image points, large gradients and these generally disrupted colour values are not taken into account during the measurement.
8. A process according to one of the preceding Claims, characterised in that the sorting of the material takes place in flight, when the material passes from one belt to another, for example.
9. A process according to one of the preceding Claims, characterised in that the measurement background is a dark hole.
10. A process according to one of Claims 1 to 8, characterised in that the measurement background is designed as a cylindrical radiator with a rotating transparent roller surrounding the radiator, wherein the radiator transmits light in a colour matched to the examination material.