JPH06102202A - Inspecting method and apparatus for woven cloth - Google Patents

Abstract

PURPOSE:To automate the inspection of the appearance quality of a cloth and to enrich the content of the appearance inspection thereby to enhance the quality of the inspection, by analyzing the frequency of image signals obtained from an inspecting cloth and studying the spatial frequency. CONSTITUTION:The reflecting light from a transferred cloth is photographed by a line sensor camera 20 to obtain video signals. The obtained video signals are sent to a shading correcting circuit, 60 for the purpose of shading correction. The video signals after the correction are developed to a two-dimensional image of one screen by a line sensor camera interface 61. The video signals of one screen are Fourier-transformed by a two-dimensional Fourier transforming device 71, and then converted to a power spectral value by a power spectrum converter 72. The value is processed through the common logarithmic operation to change the scale by a Log converter 73. The areal average value of the power spectrum of each frequency band is obtained by an areal average value circuit 74 with the use of divided areas corresponding to preset frequency bands. A classifying/rating circuit 80 outputs signals of the inspecting result of the appearance quality corresponding to the power spectrum.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a woven cloth inspection method and apparatus for inspecting the appearance of woven cloth, and more particularly to a woven cloth inspection method and apparatus for inspecting the quality of woven cloth.

[0002]

2. Description of the Related Art Conventionally, in the case of inspecting the appearance of a woven fabric, an automatic detection is performed by imaging the woven fabric by an image pickup device and comparing the brightness values obtained from the image pickup result with a threshold value to detect an abnormality in the appearance. The method is known. The content of the defect that can be detected by this automatic detection method is a defect in which the luminance value greatly changes, for example, a yarn dropout, a stain, or the like. However, quality defects called sensory defects or sensitive defects, such as the meridian muscles, the reflex muscles, and the spots, do not show an abnormal change in brightness, so it is difficult to perform physicochemical measurement, and a visual inspection is performed by the human sense. Was there.

[0003]

However, since the quality inspection by human eyes has individual differences, there is a large difference in evaluation. Further, there has been a problem in the conventional inspection method of this type that a certain degree of skill is required to perform stable evaluation.

Therefore, in view of the above points, an object of the present invention is to provide a method and an apparatus for inspecting a woven fabric, which can automatically determine the quality.

[0005]

In order to achieve such an object, the invention of claim 1 images a woven cloth to be inspected, and based on an image signal for each pixel obtained as an image pickup result, the cloth is woven. In the woven fabric inspection method for performing the appearance inspection of the cloth, when the frequency change of the magnitude of the continuous image signal obtained from the normal woven fabric is analyzed, the specific frequency obtained as a result of the frequency analysis is obtained. The power spectrum value of the belt is regarded as a characteristic indicating the quality of the woven fabric, and the allowable range of the power spectrum value which is considered to be normal in the appearance inspection of the woven fabric is set in advance, and the continuous value obtained from the woven fabric to be inspected is determined. Frequency analysis of the change in the size of the sequence of the image signal, to obtain the power spectrum value of the specific frequency band from the frequency analysis, if the obtained power spectrum value is in the allowable range Woven quality 査 subject is determined to be normal, if the acquired power spectrum value is not within the allowable range, the quality of the woven fabric of said object is characterized by determining as abnormal.

According to a second aspect of the present invention, in addition to the first aspect of the invention, the specific frequency band considered to be normal and its permissible range are defined for each of the grades of quality, and the permissible range determined is used, It is characterized in that normality / abnormality determination is executed for each type of quality.

According to a third aspect of the present invention, in addition to the second aspect of the invention, a second range of the power spectrum value corresponding to the degree of abnormality is determined for a specific frequency determined for each kind of quality, It is characterized in that the degree of abnormality is determined by comparing the power spectrum value obtained from the woven fabric to be inspected with the second range.

According to a fourth aspect of the present invention, there is provided an inspection device for picking up an image of a woven cloth to be inspected and performing an appearance inspection of the woven cloth based on an image signal for each pixel obtained as an image pickup result. When the frequency change of the magnitude of the obtained continuous image signal is subjected to frequency analysis, the power spectrum value of a specific frequency band obtained as a result of the frequency analysis is regarded as a characteristic indicating the quality of the woven fabric, and the woven fabric is obtained. In advance, the permissible range of the power spectrum value to be regarded as normal in the visual inspection is defined in advance, and frequency analysis means for frequency-analyzing the change in the size of the sequence of the continuous image signals obtained from the woven fabric to be inspected, Means for acquiring the power spectrum value of the specific frequency band from the frequency analysis, and if the acquired power spectrum value is within the allowable range, the quality of the woven fabric to be inspected is normal. Constant and, if the acquired power spectrum value is not within the allowable range, the quality of the woven fabric of said object is characterized by the fact that comprises a determination means that an abnormality.

[0009]

Function: The woven cloth has a regular structure structure composed of warp threads and weft threads. An image of such a woven fabric is represented by the crossing of light and dark waves in the warp and weft directions. For example, the reed muscle is a fine vertical stripe appearing in the longitudinal direction, and the wetting spot is a step-like defect that appears in the weft direction. According to the present invention, the inventor notices that the quality characteristic of the woven cloth appears as a change in spatial frequency, and the quality of the woven cloth is inspected based on this finding. That is, according to the first and fourth aspects of the invention, the spatial frequency is examined by frequency-analyzing the image signal obtained from the inspection object. The quality is inspected with the fact that the spatial frequency is in a specific frequency band and has a certain intensity as a normal / abnormal judgment criterion.

According to the second aspect of the present invention, by defining a normal spatial frequency band for each kind of quality such as the meridian muscles and the muscles of the reeds, it is possible to carry out an inspection of a plurality of kinds of quality and to make the inspection contents of high quality. be able to.

According to the third aspect of the present invention, when the inspection object becomes abnormal by classifying the degree of abnormality in the intensity of the spatial frequency, the abnormality content is determined.

[0012]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 2 shows the appearance of the embodiment of the present invention. In FIG. 2, a long illumination 10 such as a fluorescent lamp and a uniaxial stage 30 are provided.
The line sensor camera 20 fixed to the above is installed on the same surface of the woven cloth 40, and the reflected light is imaged. Uniaxial stage 30
Are arranged in parallel with the illumination 10, and the line sensor camera 20
Even if is moved, the change in the amount of light in the range of the field of view should be relatively small. A fluorescent lamp, an incandescent bulb, or the like is sufficient for the illumination 10 as long as it can be imaged with a relatively small amount of light,
When the transport speed is fast or the weave structure is fine, it is advisable to use a halogen light source. It is also possible to use an area sensor camera instead of the linear sensor camera 20.

The woven cloth 40 comprises two guide rolls 50, 5
1 is conveyed in the direction of the arrow. When installing the line sensor camera 20 in a normal human visual inspection machine,
It is installed so as to be along the inspection table 52. The image signal acquired by the line sensor camera 20 is image-analyzed in the main body of the inspection device of FIG. 1 to inspect its quality.

FIG. 1 shows a circuit configuration of an image processing system in the main body of the inspection device. In FIG. 1, the video signal obtained by the line sensor camera 20 is sent to the shading correction circuit 60 and is subjected to shading correction. The corrected video signal is developed into a two-dimensional one-screen image by the line sensor camera interface 61. A video signal for one screen is Fourier-transformed by a two-dimensional Fourier transformer 71 (frequency analysis means of the present invention) and then converted into a power spectrum value (intensity) by a power spectrum converter 72. More specifically, the sum of squares of the real value and the imaginary value obtained by the two-dimensional Fourier transform at the same position is obtained. Since the power spectrum is point-symmetric about the origin, the first quadrant and the second quadrant are calculated. Since the calculated power spectrum value has a wide dynamic range, Log conversion is performed to reduce it.

Next, the power spectrum value is subjected to a common logarithmic operation by a Log converter to be scale-converted and sent to an in-region average value circuit (means for obtaining the power spectrum value of a specific frequency band of the present invention) 74. . The in-region average value circuit 74 is provided with divided regions in FIG. 3A and FIG. 3B corresponding to predetermined frequency bands, and such a divided region is used to calculate the power spectrum of each frequency band. Calculate the area average value (strength). In the example of FIG. 3A, the area is divided at the same angle in the radial direction with the origin as the center, and the area average value of the power spectrum is obtained using the origins having different radii.

An example of using a neural network for the classification and rating circuit 80 is shown in FIG. In FIG. 4, the in-region average value circuit 74
The output from is used as the input of the neural network (determination means of the present invention) 90. Here, the average value of 36 regions, which is divided into 6 in the radial direction and 6 in the circumferential direction, is input. As the neural network 90, a neural network including an input layer 91, an output layer 93 and an intermediate layer 92 for performing back propagation learning is used. The neural network 90 is preliminarily learned and stored with a normal quality feature amount, a feature amount in the case of a defect, and identification information corresponding to each feature amount. In the present embodiment, the defect identification information is represented by 6 bits, and the defect type and rating are represented separately. For example, the identification information indicating (Muscle, Poor), (Muscle, Slightly Bad), (-, Good) is output. For this reason, in the case of a non-defective product, the type of defect is not necessary, and therefore the neural network 90 is made to learn only the good result of the discrimination. Further, the frequency band and its intensity range corresponding to the grades of normal and defective for each kind of quality are learned in association with the identification information. When the inspection device is configured in this way, image features collected from the inspection target, that is,
After the Fourier transformer 71 performs the frequency analysis on the features related to the spatial frequency, the quality state determination corresponding to the intensity of each frequency band, that is, the normality / abnormality determination and the abnormality degree determination for each type of quality are performed. Neural network 90
Done in.

FIG. 5 shows the positions at which an image is sampled from the woven fabric to be inspected. When the cloth width is, for example, 1000 mm and the line sensor camera has a visual field of 200 mm, the inspection in the warp direction can be eliminated by inspecting 5 places per one barb.
In the case of FIG. 5, images are taken from the left end to the right end in the drawing, and then the image pickup position is returned to the left end.

In addition to this embodiment, the following example can be carried out.

1) In the present embodiment, normal / different for each kind of quality
Although the abnormality determination and the degree of abnormality are performed, the normality / abnormality can be determined only for a specific quality. In this case (corresponding to the invention of claim 1), the intensity of the frequency band used for the inspection is selected from among the intensities for each frequency band obtained from the woven fabric to be inspected by a selection circuit such as a bandpass filter. The comparison circuit compares and determines the strength of the selected frequency band and the allowable value.

2) In this embodiment, a neural network is used as a circuit for identifying the state of quality corresponding to the power spectrum value obtained from the object to be inspected, but in addition, a comparison circuit, a lookup table, etc. may be used. it can.

3) The neural network may be divided into two, and each neural network may be made to judge whether the quality is normal or abnormal and to judge the degree of abnormality detection.

4) When recording the judgment result of the quality, it is advisable to match the inspection position of the woven cloth with the judgment result.

5) When an abnormality judgment is obtained regarding the quality, it is possible to display a warning on the display device and display the abnormal portion as an image.

6) The sampling order of the data from the inspection object may be as shown in FIG. 6 instead of FIG.

[0026]

As described above, according to the present invention,
Since the quality inspection can be performed, the contents of the appearance inspection can be abundant and the quality of the inspection can be improved.

[Brief description of drawings]

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

FIG. 2 is a perspective view showing an appearance of an embodiment of the present invention.

FIG. 3 is an explanatory diagram showing divided areas set by an in-area average value circuit 74 of FIG.

4 is a structural diagram showing a configuration of a neural network used in the classification rating circuit of FIG.

FIG. 5 is a plan view showing an inspection portion of the woven cloth.

FIG. 6 is a plan view showing another inspection portion of the woven fabric.

[Explanation of symbols]

 10 Long illumination 20 Line sensor camera 30 Uniaxial stage 40 Woven fabric 50, 51 Guide roll 52 Inspection table 60 Shading correction circuit 61 Line sensor camera interface 71 Fourier transform circuit 72 Power spectrum conversion circuit 73 Log conversion circuit 74 Area average value Circuit 90 Neural network 91 Input layer 92 Output layer 93 Intermediate layer

Claims (4)

[Claims] 1. A method for inspecting a woven cloth to be inspected, in which the appearance of the woven cloth is inspected on the basis of an image signal for each pixel obtained as an image pickup result. When a continuous change in the magnitude of the image signal is subjected to frequency analysis, the power spectrum value of a specific frequency band obtained as a result of the frequency analysis is regarded as a characteristic indicating the quality of the woven fabric, and the appearance of the woven fabric is considered. The permissible range of the power spectrum value to be considered normal in the inspection is set in advance, the change in the size of the sequence of the continuous image signals obtained from the woven fabric to be inspected is frequency analyzed, and the identification is performed from the frequency analysis. When the acquired power spectrum value is within the allowable range, the quality of the woven fabric to be inspected is determined to be normal, and the acquired power spectrum value is obtained. Fabric inspection method of the woven fabric, wherein the quality of the woven fabric of said object is to be determined as abnormal if the value is not in the allowable range. 2. The specific frequency band considered to be normal for each type of quality and its allowable range are determined, and normality / abnormality determination is performed for each type of quality using the determined allowable range. The method for detecting a woven fabric according to claim 1, wherein: 3. A power spectrum value obtained from the woven fabric to be inspected, which defines a second range of power spectrum values corresponding to the degree of abnormality for specific frequencies defined for each type of quality, The method for inspecting a woven fabric according to claim 2, wherein the degree of abnormality is determined by comparing with the second range. 4. An inspecting device that images a woven fabric to be inspected and inspects the appearance of the woven fabric on the basis of an image signal for each pixel obtained as an imaging result. When the frequency change of the magnitude of the image signal is subjected to frequency analysis, the power spectrum value of a specific frequency band obtained as a result of the frequency analysis is regarded as a characteristic indicating the quality of the woven fabric, and is normally detected in the appearance inspection of the woven fabric. The permissible range of the power spectrum value to be regarded as is determined in advance, and frequency analysis means for frequency-analyzing a change in the size of a row of the continuous image signals obtained from the woven fabric to be inspected, A means for acquiring the power spectrum value of a specific frequency band, and when the acquired power spectrum value is within the allowable range, it is determined that the quality of the inspection target woven fabric is normal, and the acquisition is performed. If the power spectrum value does not fall within the allowable range, the inspection device further comprises a determination unit that determines that the quality of the inspection target woven fabric is abnormal.