EP0852709A1

EP0852709A1 - Multispectral sensor device

Info

Publication number: EP0852709A1
Application number: EP95933433A
Authority: EP
Inventors: Norbert Haase; Andreas Kalz; Jens Knobloch
Original assignee: Fraunhofer Gesellschaft zur Forderung der Angewandten Forschung eV
Current assignee: Fraunhofer Gesellschaft zur Forderung der Angewandten Forschung eV
Priority date: 1995-09-27
Filing date: 1995-09-27
Publication date: 1998-07-15
Also published as: WO1997012212A1; US5926282A

Abstract

A multispectral sensor device comprises a plurality of optoelectrical transducers (1), each of which wavelength-selectively generates from an optical signal an electrical measurement signal in its own measurement channel. A processing circuit (2, 3, 4) processes each electrical measurement signal to produce measurement signal values. A fuzzy-logic circuit (6) performs a comparison of the measured signal values with reference values and on the basis of that comparison allocates measurement signal values to a number of channels (7), that number being greater than the number of measurement channels.

Description

Multispectral sensor device

description

The present invention relates to a multispectral sensor device for analyzing optical radiation with an unknown spectral composition and intensity distribution, predominantly from the ultraviolet (UV) to the visible (VIS) to the near infrared (NIR) range.

Such a sensor is, for example, as a color detection system with the predetermined three spectral components red, green and blue in the VIS range, as an extended color detection system with the additional spectral components as color equivalents, or when increasing the number of channels and the associated reduction of the spectral bandwidth can also be used as a channel spectrometer.

Various color measurement systems or spectral analyzers are known in the art.

DE 36 22 043 AI describes a device for color measurement. This device requires at least two predefined spectral sensitivity functions and for spectral filtering a multiplicity of illuminated filters, a detector arrangement with one detector behind each filter and an evaluation circuit which receives the detector signals emitted by the detectors and the measured value signals assigned to the individual sensitivity functions ab¬ there. A plurality of weighting factors, each assigned to the individual sensitivity curves, are stored in a data memory of the evaluation circuit. The evaluation circuit derives the measured value signals, which are assigned to the individual sensitivity curves, from the detector signals of these detectors, the same with those of the weighting factors assigned to the respective sensitivity function. With this arrangement, only a rough color determination can be carried out. For spectral analyzes with, for example, 10 nm resolution per measuring channel, 40 channels would be required in the visible range (380 nm to 700 nm), which can only be implemented in this way with great effort.

DE 41 43 284 AI describes an integrated semiconductor sensor for spectrometers. This sensor consists of a large number of radiation detectors, charge storage and charge transport devices, supply potential lines, logic circuits and output amplifiers. These devices are arranged in a one- or two-dimensional arrangement in order to convert the radiation from partial areas of the spectrum into electrical signals. The radiation detectors arranged on one or more contiguous wavelength ranges of the spectrum imaged on the integrated semiconductor sensor each have dimensions which are adapted to the spatially resolving conditions which result from the aperture and dispersion system of the spectrometer and the radiation source to be examined and are corresponding to the Row structure of the spectrum arranged. The storage charge transport amplifier and circuit devices connected downstream of the detectors are set up in a manner which is specifically adapted to the sensors used in each case. By means of such an arrangement together with an integrated logic circuit, the signals of the detectors of selectable connected sections of at least one wavelength region regularly occupied by detectors are combined individually or partially and read out serially and / or in parallel. Furthermore, the light integration time of the individual sections is controlled.

This sensor can also be used as a color sensor. However, its disadvantage is that each sub-area of the spectrum requires a correspondingly adapted detector in the measuring channel. This means that either the spectral limited range or a large number of such detectors must be implemented to capture the entire spectral range. In the case of many measurement problems, however, this is not possible since the measurement beam of the incident light must be optically widened in such a way that all spectrally adapted detectors are irradiated in the same way. As a result, the actual semiconductor sensor becomes large and the luminance at the individual detector drops.

DE 41 33 581 AI describes another multispectral sensor. In this, the measuring beam is broken down into several partial beams by an optical device and reflected on filters. Radiation-sensitive elements are arranged behind the filters, which detect the spectral range that corresponds to the transmission range of the upstream filter. The remaining spectral region of the partial beam is reflected from the filter surface onto the other filters. Such an arrangement only makes sense for a few spectral ranges, since otherwise this sensor also requires a large and complicated optical structure.

Another color sensor is disclosed in the document "Sensor and Actuators A" 41-42 (1194), pages 123-128. For the determination of the red, green and blue components of the radiation in the visible spectral range, vertically stacked silicon diodes are used, by means of which the wavelength-dependent absorption of the visible light in silicon is used to set the required spectral ranges. This solution is elegant for determining the color components of a spectrum, since the dimensions of such a sensor can be kept small. However, the spectral bandwidth of the three spectral ranges and their center of gravity can only be selected to a limited extent because of the technological depth of the detection volumes of the pn junctions. As a result, the sensor cannot be used as a multispectral sensor.

The known multispectral sensors for the UV-VIS-NIR spectral range have the disadvantage that they are used for each measuring channel a detector and / or, for spectral decomposition, additional optical paths and components, such as gratings or prisms, which are sometimes also arranged to be movable, are required, as a result of which they are correspondingly large and complex to manufacture.

Starting from the prior art mentioned, the object of the present invention is to create a multispectral sensor device which, by detecting a few spectral subregions with a relatively large spectral bandwidth, enables statements to be made about a larger number of spectral subregions with a smaller bandwidth .

This object is achieved by a multispectral sensor according to claim 1.

The present invention provides a multispectral sensor device with the following features: a plurality of opto-electrical transducers, each optoelectric transducer generating wavelength-selectively from an optical signal an electrical measurement signal in a respective measurement channel; a processing circuit in which the electrical measurement signals generated are each processed into measurement signal values; and a fuzzy logic circuit in which the measurement signal values generated are compared with reference values and the measurement signal values are assigned on the basis of the comparison to a number of channels which is greater than the number of measurement channels.

The optoelectrical transducers used in the present invention are preferably photodiodes, which are each provided with multilayer filters which are spectrally different from one another. The processing circuit preferably has measurement amplifiers, analog / digital converters and measurement value accumulators in order to convert the measurement signals into measurement signal values. The measurement signal values are then fed to the fuzzy logic circuit via a reference value memory in which the reference values are stored. Such a sensor can be integrated together with analog and digital signal processing components as a microsystem on a chip in an extremely small monolithic manner and can therefore be produced inexpensively in large quantities. Very high system requirements can also be met, in particular because of the fuzzy logic circuit of the detector system. In addition, extreme miniaturization is achieved in comparison to previous spectrometer systems, since the number of detector channels required is reduced to a minimum, while the need for optical paths for spectral decomposition of the light and for moving parts, such as gratings or Prisms, and movable detectors is eliminated.

The fuzzy classification of the measurement signal values enables a sufficiently reliable evaluation even in the case of inexactly defined (unsharp) ambient conditions, such as, for example, the lighting and the associated evaluation of the measured material or in the case of disturbed signals. The unfavorable properties of inexpensive integrated sensors compared to expensive individual sensors, which are due to unavoidable tolerances in the joint production of all photodiodes and multilayer filters and to the lack of individual optimization of each sensor component, are compensated for by suitable design of the fuzzy logic circuit.

According to the present invention, a multispectral sensor device with, for example, n-channels for the UV-VIS-NIR range is created which, by means of the acquisition of only a few spectral subregions with a relatively large spectral bandwidth, has a large number of spectral ranges by means of integrated fuzzy logic Sub-areas generated significantly lower bandwidth. These are assigned to discrete channels, so that this sensor device can be used both for determining the color components or color equivalents and for discrete spectral analysis with the channel width as a discretization measure, the discretization measure corresponding to the spectral resolution of the multispectral sensor device. By means of a Such multispectral sensor devices can be used, for example, to make statements regarding the classification of the deviation with respect to a predetermined color, the occurrence of defined spectral components or the color of the measurement object.

Such a sensor can be used in many ways, especially since it can be produced inexpensively and in large quantities as a monolithically integrated component, for example in Si-CMOS technology, together with analog and digital signal processing units on a chip.

Preferred exemplary embodiments of the present invention are explained below with reference to the accompanying drawings. Show it:

1 shows an embodiment of the multispectral sensor device according to the present invention; and

2 shows the course of the spectral sensitivity of the five photodiodes described in the exemplary embodiment.

As shown in FIG. 1, a preferred exemplary embodiment of the present invention has five photodiodes 1 (UVA, B, G, R, NIR). These photodiodes are applied to a silicon substrate and each are provided with a bandwidth and center wavelength desired for the technology and suitable multilayer filter. The multilayer filter or filter layer system is applied to the photoactive region of the diode, which is located in a thin semiconductor layer, an insulating intermediate layer being arranged between this semiconductor layer and the Si substrate.

Each of the diodes is connected to a measuring amplifier 2. Each measuring amplifier is in turn connected to an analog / digital converter 3. The analog / digital converters are connected to accumulators 4, the outputs of which are connected to a measuring and Reference value memory 5 are connected. The outputs of the measurement and reference value memory 5 are connected to a fuzzy logic unit 6. If necessary, temperature compensation can also be added to this circuit.

If radiation of any spectral composition and intensity distribution falls on the multispectral sensor, it is detected by its five photodiodes simultaneously, but according to their spectrally different sensitivity. The detected measurement signals are fed to the respectively assigned measurement amplifiers, the output signals of the same are converted into digital signals by analog / digital converters and, after a digital characteristic curve correction, are integrated in an accumulator. The measurement signal values thus obtained are then fed to the fuzzy logic circuit via the measurement and reference value memory 5. The fuzzy logic circuit has the task, among other things, of forming the spectral components of the light to be measured with a higher resolution from the signals of a few photosensors (in the present example 5) and realizing a classification. Different fuzzy systems are provided for different applications. First, the comparison with a given color, i.e. a classification of the deviation with respect to the specified color, secondly a spectral analysis, i.e. a fuzzy statement about the occurrence of defined spectral components, and thirdly a color classification, i.e. a fuzzy statement about the color of the measurement object. Furthermore, object detection and classification can be carried out by means of the fuzzy logic circuit 6.

FIG. 2 shows the course of the spectral sensitivity of the 5 diodes used in the preferred exemplary embodiment. In this exemplary embodiment, the multilayer filters of the five diodes are dimensioned for the spectral range UV-VIS-NIR (250 nm to 950 nm). The dimensioning of the five multilayer filters or filter layer systems for the respective diodes is described below. 1st diode (UVA):

The filter layer system of the 1st diode is due to the successive application of SiO _{2 with} a thickness of 320 nm to 340 nm, Al with a thickness of 10 nm to 15 nm, SiO _{2 with} a thickness of 180 nm, Al with a thickness of 10 nm to 15 nm and finally SiO 2 with a thickness of 80 nm to 90 nm in the photoactive region of the diode. The filter dimensioned in this way has a maximum transmission 8 at a wavelength of around 290 nm with a half width of approximately 55 nm and has a maximum transmission of approximately 65%.

2nd diode (VIS blue):

The filter layer system of the second diode is formed by the successive application of TiO _{2 with} a thickness of 50 nm, SiO _{2 with} a thickness of 125 nm and finally Al with a thickness of 10 to 15 nm in the photoactive region of the diode. The filter dimensioned in this way has a maximum transmission 9 at a wavelength of approximately 440 nm with a half width of approximately 75 nm. The maximum transmission 9 is approximately 70%.

3rd diode (VIS green):

The filter layer system of the third diode is formed by the successive application of TiO _{2 with} a thickness of 50 nm, SiO _{2 with} a thickness of 160 nm and Au with a thickness of 30 nm to 35 nm in the photoactive region of the diode. This filter has a maximum transmission 10 at a wavelength around 550 nm with a half width of 130 nm. The maximum transmission 10 is approximately 65%.

4. Diode (VIS red):

The filter layer system of the fourth diode is formed by the successive application of TiO _{2 with} a thickness of 50 nm, SiO _{2 with} a thickness of 220 nm and Au with a thickness of 30 to 35 nm in the photoactive region of the diode. The maximum transmission 11 at a wavelength around 680 nm has a half width of approximately 115 nm and a maximum value of about 60%.

5. Diode (NiR):

The filter layer system of the 5th diode is photoactive by successively applying TiO _{2 with} a thickness of 50 nm, SiO _{2 with} a thickness of 560 nm to 580 nm and polycrystalline silicon with a thickness of 1.5 μm to 2.1 μm Be¬ area formed. The maximum transmission 12 of this filter at approximately 810 nm has a half width of approximately 240 nm. The maximum transmission is about 55%.

Alternatively, optoelectric transducers can be used to acquire other spectral ranges, which receive a correspondingly differently adapted spectral sensitivity.

The fuzzy logic circuit of the present invention will now be described.

The fuzzy logic circuit extracts features from the measurement signal values and adds further application-specific ones. Using the membership functions and fuzzy rules stored in the fuzzy system, all features are overlaid with the aim of optimal classification. The measured signal values of, for example, five photodiodes are assigned, for example, to n spectral channels with an association between 0 and 1. It is thereby achieved that a high discrete spectral resolution is achieved with only a few photodiodes which differ in their spectral sensitivity. The fuzzy system is adaptive by means of appropriate algorithms and storage methods, i.e. it can be optimized and adapted to changing conditions.

A fuzzy logic that can be used in the fuzzy logic circuit of the preferred exemplary embodiment is described below.

To set up the color classifier with fuzzy logic, it is not necessary to to describe thematically or to determine the statistical distribution functions. At the beginning, the measurement signal values detected by series of measurements by means of the photodiodes are considered. These measured signal values are compressed and processed into diagnostic features, which together form the feature vector for the system under consideration. The "unsharpness" of production and measurement are taken into account.

As shown in Figure 2, the feature vector in the present example consists of the five values "UV", "blue", "green", "red" and "NIR", which can be supplemented or reduced in specific applications. The signal size of each feature becomes value ranges, e.g. "close to zero (nn)", "small (k)", "medium (m)", "medium (mg)", "large (g)" and "very large (sg)", which are fuzzy are limited. Corresponding empirical values are used for the assignment to a signal variable, this assignment forming the membership function of the corresponding feature.

A connection, which is defined by verbal fuzzy rules, to the feature vector is formed for the desired classes (for example 8 special colors). If, for example, a class "blue-green" is to be formed according to Figure 2, one possible rule is:

IF

^« UV" = "nn" and "blue" = "mg" and "green" = "g" and "red" =

"k" and "IR" = "nn"

THEN "teal"

Further rules for this class would improve the clarity of the information. The feature vector, the corresponding membership functions and the entire set of fuzzy rules form the basis of the classifier. If there are n classes, the classifier forms an n-dimensional feature space in which each class corresponds to a sub-area. The task of this fuzzy classifier is now to assign an object, in this case a color, to a class, even if this signal is disturbed by additional information, for example a fluctuating luminosity, a gloss or the surface structure. Since the individual classes are described in a vague manner, the result is not a sharp value, but rather a certain degree of belonging to a class. This degree of belonging, also called sympathy, can take values between 0 and 1. If, for example, the above rule is fully fulfilled, the degree of affiliation or sympathy for the class "blue-green" has the value 1. A reduced luminosity or results of further rules for the class "blue-green" can of course reduce this value.

With a suitable optimization of the classifier, this method makes it possible to define color levels between the measured color values and to interpret transitions. In the classification decision based on the evaluation of the sympathy values of each class, three cases are to be distinguished:

1. The sympathy value of a class is significantly larger than all other sympathy values: the object is clearly assigned to a class.

2. The sympathy value of a class is only marginally greater than another sympathy value: the classification decision can only be accepted with reservations, and there may be a tendency to an adjacent class.

3. All sympathy values are less than a given identification threshold: the object cannot be assigned to a class. With the fuzzy evaluation of fewer sensor signals selected according to the preferred exemplary embodiment, an increase in the number of color classes is possible. By defining the necessary sympathy values and identification thresholds, a high level of information security is possible even in poorly defined environmental conditions.

The mentioned properties of the multispectral sensor device with fuzzy logic of the present invention enable the same to be used for all tasks which relate to a real or false color analysis. These include tasks of colorimetry in the color, automotive and printing industries, as well as color rendering techniques. The present invention can also be used in multispectral image analysis, for example remote earth detection, soil analysis, environmental protection and medical technology. Furthermore, the multispectral sensor device according to the present invention can be used for three-dimensional object recognition by means of color triangulation. Further areas of application are sorting systems and devices in which artificial vision is required.

Another area of application is spectrometry, e.g. for reaction kinetic measurements, for determining the rate constants of chemical reactions and for clarifying the reaction mechanisms. Such measurements require a fast sensor which is capable of detecting a larger wavelength range selectively in situ in a very short time.

Furthermore, the present invention can be used in the paper and printing industry, in which high demands are placed on a multispectral sensor device. A basic prerequisite for on-line color measurement in the paper manufacturing and printing process is that the same must take place without contact, and that the material to be measured is fluttering at the same speeds because of it is difficult to fix. In the case of lighter papers, the problem of translucency, which has a fluctuation in the measurement results, the cause of which lies in the local paper structure, makes matters more difficult. The multispectral sensor device according to the present invention has all the features in order to be able to carry out such measurements satisfactorily.

Claims

claims

1. Multi-spectral sensor device with the following features:

a plurality of optoelectric transducers (1), each optoelectric transducer generating a wavelength-selective electrical signal from an optical signal (8 to 12) in each measuring channel; and

a processing circuit (2, 3, 4) in which the generated electrical measurement signals are each processed in measurement signal values; marked by

a fuzzy logic circuit (6) which compares the generated measurement signal values with reference values and assigns the measurement signal values on the basis of the comparison to a number of channels (7) which is larger than the number of measurement channels.

2. Multi-spectral sensor device according to claim 1, characterized by

that the optoelectric converters are photodiodes, pin diodes, Schottky diodes or MOS structures, which are each provided with spectrally different multilayer filters.

3. Multi-spectral sensor device according to claim 1 or 2, characterized in that

that the processing circuit for each measuring channel has a measuring amplifier (2), an analog / digital converter (3) and a measuring value accumulator (4), by means of which the individual measuring signals are processed to measuring signal values, the measuring signal values being stored in a reference value memory (5) , in which the reference values are stored, are fed to the fuzzy logic circuit (6). 4. Multi-spectral sensor device according to one of claims 1 to 3, characterized in that

that the multispectral sensor (1) and the processing circuit (2, 3,

4) are integrated as a microsystem on a chip.

5. Multi-spectral sensor device according to one of claims 1 to 4, characterized in that

that the fuzzy logic circuit (6) compares the measurement signal values with predetermined spectral colors, which are represented by the reference values, and carries out a classification of the color deviation.

6. Multi-spectral sensor device according to one of claims 1 to 4, characterized in that

that the fuzzy logic circuit (6) carries out a spectral analysis with respect to the spectral composition and intensity distribution of the optical radiation impinging on the multispectral sensor from the measurement signal values of each measurement channel and makes a statement about the occurrence of defined spectral components.

7. Multi-spectral sensor device according to one of claims 1 to 4, characterized in that

that a color classification of the test object is carried out by means of the fuzzy logic circuit (6).

8. Multi-spectral sensor device according to one of claims 1 to 4, characterized in that

that object detection and classification is carried out by means of the fuzzy logic circuit (6).