JPH06118017A - Quality inspecting device - Google Patents

Abstract

PURPOSE:To precisely inspect the quality of a color-changing surface by quantizing a video signal to conduct differential processing in the color changing direction, the result vertically to the color changing direction, then performing Fourier transformation. CONSTITUTION:A film 10 is exposed to an illumination 11, and the transmitted light is picked up by a television camera 12 and quantized by an A/D converter 21 to form a digital image, which is then inputted to an image memory 22. On the basis of this image, the quality of a color transitional part on the surface is judged by a CPU 23 and transmitted from an output part 30 to the outside. To the image data, a spatial differential value in N-direction (color transitional direction) is determined, each signal in N-direction is integrated in M-direction, and the average density in M-direction is determined with a specified expression to form an one-dimensional signal. This signal is subjected to hamming window processing to remove the influence of edge, and a power spectrum is obtained by Fourier transformation. This power spectrum is inputted to a neural network to judge its rank.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quality inspection device for inspecting the quality of the color change state of the surface of a film or the like.

[0002]

2. Description of the Related Art Conventionally, there is a method of inspecting for unevenness occurring on a uniform surface by dividing an image pickup area into small areas and comparing the uniformity with neighboring areas (Japanese Patent Laid-Open No. 63-2.
00278).

[0003]

However, the prior art has the following problems. It is not possible to inspect the state of the object to be inspected, which does not have uniformity such as color change.

It is impossible to make a delicate rank determination that matches a human sense only by comparing with a simple threshold value.

The present invention determines the quality condition of a surface that changes color.
It is an object of the present invention to provide a quality inspection device that conforms to human sensitivity and inspects with high accuracy and certainty.

[0006]

SUMMARY OF THE INVENTION The present invention is a quality inspection apparatus for imaging a surface of an object to be inspected and determining the quality state of the color change by a neural network, wherein a video signal from the inspection object is detected. Quantize and perform differential processing in the color change direction, integrate each signal in the vertical direction in the color change vertical direction, and perform Fourier transform on this to obtain its power spectrum, and to convert them to the neural network. The input signal is used.

However, the description of the "neural network" in the present invention is as follows (1) to (4).

(1) The neural network has a structure similar to that of the hierarchical network shown in FIG.
It can be roughly classified into two types of mutual coupling networks shown in (B). The present invention may be configured by using either of both networks, but the hierarchical network is more useful because a simple learning algorithm as described later has been established.

(2) Network Structure As shown in FIG. 7, the hierarchical network has a hierarchical structure including an input layer, an intermediate layer, and an output layer. Each layer is composed of one or more units. The coupling is only forward coupling such as input layer → middle layer → output layer, and there is no coupling within each layer.

(3) Structure of the unit The unit is a model of a brain neuron as shown in FIG. 8 and has a simple structure. Receive input from other units,
The sum is taken and converted by a certain rule (conversion function), and the result is output. A variable weight that represents the strength of the connection is attached to each of the connections with other units.

(4) Learning (Back Propagation) Learning a network is to bring an actual output close to a target value (desired output). Generally, the conversion function and weight of each unit shown in FIG. Is learned by changing.

As a learning algorithm, for example, Rumelhart, DE, McClelland, JL and the PDP Re
search Group, PARALLEL DISTRIBUTED PROCESSING, the
Backpropagation described in MIT Press, 1986. can be used.

[0013]

When the quality state of the color change on the surface of the object to be inspected is judged by the neural network, the video signal from the inspection object is quantized and the differential processing is performed in the color change direction, and each signal in the vertical direction is colored. Since it integrates in the vertical direction of change and uses the above-mentioned integration result as input data to the neural network, it is necessary to accurately and accurately inspect the quality condition of the surface that changes color in conformity with human sensitivity. You can

[0014]

FIG. 1 is a block diagram showing an embodiment of the present invention,
2 is a flow chart showing an inspection procedure, FIG. 3 is a schematic diagram illustrating a target pixel at the time of differentiation, FIG. 4 is a schematic diagram showing an example of a differential operator, FIG. 5 is a schematic diagram showing Hamming window processing,
6 is a schematic diagram showing a neural network, FIG. 7 is a schematic diagram showing a hierarchical neural network, and FIG. 8 is a schematic diagram showing the structure of a unit.

FIG. 1 is a block diagram of an apparatus for inspecting a shaded film for laminated glass. This film is semi-transparent because it has extremely minute irregularities on the surface, and is partially colored, and the color gradually becomes lighter in one direction and becomes colorless. In the present invention, the changing state of the color transition portion is inspected.

Illumination 11 is applied from below to the film 10 which is the object to be inspected, and the transmitted light is transmitted to the television camera 1.
The image is captured by 2.

In the A / D converter 21, for example, 8 bits (2
Quantization is performed with 56 gradations to create a digital image of M * N pixels, and the digital image is input to the image memory 22.

Based on the input image, the CPU 23
The quality of the color transition portion on the surface is determined by and is transmitted from the output unit 30 to the outside. FIG. 2 shows the flow until quality judgment.

For the image data of M * N pixels, a spatial differential value in the N direction (corresponding to the color transition direction) (actually a signal is digitized and thus a difference value) is obtained. Here, the density at coordinates (i, j) as shown in Fig. 3
n * 3 with the differential value g (i, j) of f (i, j) centered on coordinates (i, j)
Calculated from data in the range. That is, for example, n = 5 and using the operator shown in FIG. 4, g (i, j) =-f (i-2, j-1) -f (i-1, j-1) -f (i, j-1) -f (i + 1, j-1)-
f (i + 2, j-1) + f (i-2, j + 1) + f (i-1, j + 1) + f (i, j + 1) + f (i + 1,
Let j + 1) + f (i + 2, j + 1).

Each signal in the N direction is integrated in the M direction. Since n = 5 here,

[Equation 1] Then, the average density in the M direction is calculated to obtain a one-dimensional signal.

A Hamming window process (see FIG. 5) is applied to the one-dimensional signal to remove the influence of edges, and a power spectrum is obtained by Fourier transform. Where the Hamming window function is

[Equation 2] Is defined by

This power spectrum is input to the neural network to determine its rank. The neural network is a three-layer hierarchical network (see FIG. 7), the input layer has the same number of units as the number of points of the power spectrum, and the output layer has the same number of ranks to be determined. This network is preliminarily learned by back propagation so that the output unit corresponding to the rank with respect to the input is 1, and the others are 0.

In the practice of the present invention, it is obvious that the same result can be obtained even if a line sensor is used for imaging the surface. Also, n of the differential operator is arbitrary,
For the subsequent integration, it is not always necessary to use the full range of differentiation.

[0024]

As described above, according to the present invention, it is possible to reliably and accurately inspect the quality condition of the surface where the color changes, in conformity with human sensitivity.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a flow chart showing an inspection procedure.

FIG. 3 is a schematic diagram illustrating a target pixel at the time of differentiation.

FIG. 4 is a schematic diagram showing an example of a differential operator.

FIG. 5 is a schematic diagram showing a Hamming window process.

FIG. 6 is a schematic diagram showing a neural network.

FIG. 7 is a schematic diagram showing a hierarchical neural network.

FIG. 8 is a schematic diagram showing a structure of a unit.

[Explanation of symbols]

 10 film (inspection object) 11 illumination 12 TV camera 21 A / D converter 22 image memory 23 CPU 30 output unit

Claims (1)

[Claims] 1. A quality inspection apparatus for imaging a surface of an object to be inspected and determining the quality state of the color change by a neural network, wherein the video signal from the inspection object is quantized and differentiated in the color change direction. It is characterized in that processing is performed, each signal in the vertical direction is integrated in the color change vertical direction, and the Fourier spectrum is applied to this to obtain the power spectrum, and these are used as input signals to the neural network. Quality inspection equipment.