CN1640331A - Lines feature on-line identifying method for detecting foreign matter in food - Google Patents

Abstract

The present invention relates to an on-line identification method for detecting foreign matter in article by utilizing grain chracteristics. Said method includes the following steps: (1). shooting image; (2). image transmission; (3). foreign matter identification processing; (4). identification signal transmission; and (5). removing foreign matter. Said invention adopts LED as background light source, and uses the colour close the colour of article as background colour and adopts FFT to make grain analysis of the image so as to raise correct identification rate of articles of tobacco, etc. and foreign matter.

Description

The textural characteristics ONLINE RECOGNITION method that foreign matter detects in the article

Technical field:

The present invention relates in the online detection of a kind of article the method for discerning than the foreign matter of obvious textural characteristics is arranged.The foreign matter that is mainly used in article such as tobacco detects.

Background technology:

At present, the conventional foreign matter recognition methods of using in article on-line detecting systems such as tobacco is based on the colourity statistic law of pixel, its basic principle is to utilize tobacco and the foreign matter difference on color, pixel belongs to tobacco or foreign matter in the judging unit, and then whether includes foreign matter in the judging unit.Its course of work has following basic step: 1, chroma space: transform to Ohta or HIS space by rgb space; 2, computational discrimination threshold value: the scope (as R, G, B etc.) of the color parameter of statistics tobacco leaf and various foreign matters in chrominance space, the upper and lower bound that obtains differentiating; 3, the pixel attribute is differentiated: according to the threshold value that obtains, the pixel of differentiating in the testing image belongs to tobacco leaf or foreign matter; 4, cell attribute is differentiated: the ratio that belongs to tobacco leaf according to pixel in the unit determines whether this unit comprises foreign matter.

Said method has recognition effect preferably for the bigger foreign matter of color and tobacco leaf difference, but powerless for the color foreign matter similar to tobacco leaf.In the step (3) of said method, be that to differentiate this point be to belong to tobacco or foreign matter for colorimetric parameter (as R, G, B etc.) according to pixel.In the actual conditions, some foreign matter has in a big way overlapping with tobacco in color controls, therefore only differentiate according to color of pixel, is easy to occur erroneous judgement.This also is the place that this method haves much room for improvement.

Summary of the invention:

In order to improve the correct decision rate of article such as tobacco and foreign matter, especially improve the obviously discrimination of the foreign matter of textural characteristics is arranged, to develop the more excellent tobacco eliminating system of performance, the invention provides the textural characteristics ONLINE RECOGNITION method that foreign matter detects in a kind of article.It has utilized standard article and the difference of foreign matter on texture, and through the FFT conversion, the transform domain image of standard article and foreign matter has tangible different, by contrasting the transform domain image of the two, can judge that it is foreign matter or article.

In order to realize the foregoing invention purpose, technical scheme of the present invention is:

The present invention has comprised the basic step of prior art:

(1) image capture:

(2) image transmission:

(3) foreign matter identification is handled:

(4) identification signal transmission:

(5) reject foreign matter.

Improvement of the present invention is:

In step (3),, utilize based on whether including foreign matter in the pixel colourity diagnostic method recognition unit at first according to the colour information of image; If can't utilize the colourity of pixel to judge, then carry out texture recognition according to the FFT conversion of unit.Why to adopt earlier based on pixel colourity diagnostic method and discern, be because FFT transform operation amount is bigger, than in the background technology based on the method complexity of pixel colourity, and the color unit proportion similar to the standard article can be very not big generally speaking, in order to shorten recognition time under the prerequisite that improves recognition effect as far as possible, this method need cooperate based on the method for pixel color degree to be used.Running into the color unit similar, can't the judgement unit attribute time, just using the FFT conversion to carry out texture recognition and use based on the method for pixel colourity to the standard article.

Utilizing the FFT conversion to carry out texture analysis, is that to be reflected in the FFT changing image according to different textures be significantly speck to occur in certain location, the identification that position that occurs by the contrast speck and rule realize foreign matter.The basic principle of this method is as follows:

One width of cloth digital picture can be thought the spatial domain discrete series function of a two dimension.With f (i, k) expression one width of cloth digital picture, then the form of its discrete Fourier transform is:

$F(u,v)=\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y)e^{-j2\mathrm{\π}(\frac{\mathrm{ux}}{M}+\frac{\mathrm{vy}}{N})}$

In the formula, u=1,2 ..., M-1; V=1,2 ..., N-1.

In the time of Practical Calculation, make usually image be the square, this moment M=N.And, utilize the separability of two dimensional discrete Fourier transform, it is become two one-dimensional transforms calculate.The separability of two dimensional discrete Fourier transform is shown below:

$F(u,v)=\frac{1}{N}\underset{x=0}{\overset{N-1}{\mathrm{\Σ}}}{e}^{-j2\mathrm{\π}\frac{\mathrm{ux}}{N}}\×\underset{x=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y)e^{-j2\mathrm{\π}\frac{\mathrm{ux}}{N}}----u,v=\mathrm{1,2},\·\·\·,N-1$

FFT is the fast algorithm of Fourier transformation, can be divided into decimation in time method and decimation in frequency method two big classes, and this is a prior art, does not introduce in detail herein.

Among the present invention, employing may further comprise the steps based on the recognition methods that FFT carries out texture analysis:

1. gray processing: the coloured image of standard article is transformed to gray level image;

2. smothing filtering: image is carried out medium filtering, eliminate noise.Specifically be on the gray level image of gained, to use a sliding window, the gray scale of each pixel in the window is sorted by size, replace the former gray scale of window center pixel with its intermediate value;

3. FFT conversion: several standard images of items and background image are carried out the FFT conversion, the view data after the conversion is deposited in the internal memory; Treat detected image and carry out the FFT conversion, obtain transform domain image;

4. discern according to changing image: the FFT changing image of testing image and standard article and background is compared.Concrete grammar is as follows: at first make the two center brightness normalization, equate even two amplitude variations to be distinguished change the area image center brightness, make simultaneously that the ratio of each pixel and central pixel point brightness remains unchanged in the image separately; Be the center then with the central pixel point, the length of respectively getting certain pixel quantity up and down marks off a rectangle, and in the part in the statistical picture outside rectangle, brightness surpasses certain value G ₁The number N of pixel; Several standard article in the internal memory and background image are repeated aforesaid operations, with the maximum N of the N of several transform domain images _MaxAs the upper limit, when the N of testing image greater than N _Max, then can regard as the image that includes foreign matter, otherwise then be images of items.

Further, the present invention also in step (1), uses light emitting diode to replace common fluorescent tube as light source, because lifetime of LED is long, stabilized intensity, the color and the brightness that obtain image are also just more stable.In order to discern common foreign matter with simpler method (as diagnostic method) based on pixel colourity, adopted the color close look as a setting with article, strengthen the aberration of background and foreign matter.With two CCD line array video cameras the upper and lower surface of determinand is taken pictures simultaneously, respectively two width of cloth images that obtain are discerned, reduced like this owing to block the erroneous judgement of generation.

The invention has the beneficial effects as follows: some foreign matter colourities are similar to article to be measured, and a large amount of coincidences zone is arranged in the color space, only use in the background technology based on the recognition methods of pixel colourity, and discrimination is very low, and higher False Rate is arranged; If foreign matter has significantly textural characteristics, then can be by method of the present invention, employing FFT carries out the textural characteristics analysis, discerns with the image after the conversion, thereby has improved the correct recognition rata of article such as tobacco and foreign matter.Although the operand of FFT is bigger, but many digital signal processing chips (DSP) are all integrated FFT calculation function, therefore by embedding DSP, realize the FFT conversion with hardware-efficient, the method among the present invention also can be used for the real-time processing of the online detections of article such as tobacco easily.

Description of drawings:

Fig. 1 is the detection system structural representation based on textural characteristics ONLINE RECOGNITION method.

Fig. 2 is the flow chart of textural characteristics ONLINE RECOGNITION method.

Fig. 3 is the original image of standard tobacco leaf.

Fig. 4 is the original image of Calusena lansium band.

Fig. 5 is the gray level image of standard tobacco leaf.

Fig. 6 is the FFT changing image of standard tobacco leaf.

Fig. 7 is the gray level image of Calusena lansium band.

Fig. 8 is the FFT changing image of Calusena lansium band.

Specific embodiment:

Below with the example that is identified as of tobacco, in conjunction with the accompanying drawings, a specific embodiment of the present invention is described.

In conjunction with Fig. 1 and Fig. 2, being implemented as follows of this method:

1, image capture: on production line, when tobacco leaf stream is transferred into the conveyer belt rear end, produce high velocity air with starting drive, tobacco leaf stream is coupled in the air-flow and with high speed (about 5m/s) impelling, flow when unsettled at tobacco leaf, (5000 lines/s) absorb the tobacco in the certain limit and the coloured image of foreign matter mixture simultaneously from upper and lower surface to adopt two high-speed CCD line array video cameras.Light source adopts and replaces common fluorescent tube with light emitting diode.

2, the image that the CCD line array video camera is obtained reaches in the DSP embedded system and handles, and the position signalling of tobacco leaf stream is reached in the middle of the calculator memory.

3, foreign matter identification is handled:

At first discern, can't discern under the situation whether foreign matter is arranged, adopt method of the present invention in this method with the method based on pixel colourity in the background technology:

(1) to the original color image gray processing of standard tobacco leaf shown in Fig. 3 and Fig. 4 and Calusena lansium band, makes it to become gray level image such as Fig. 5 and shown in Figure 7.

(2) gray level image is carried out medium filtering, reduce influence of noise.

(3) carry out the FFT conversion, obtain transform domain image, as Fig. 6 and shown in Figure 8.

(4) by in the Fourier transformation image of contrast discovery tobacco leaf (Fig. 6), the center speck is very outstanding, and the darker clear zone of periphery lacks tangible directionality, and concentrates on around the central bright spot; And the Fourier transform image (Fig. 8) of Calusena lansium band, central bright spot is slightly little dark slightly, but has more brightness bigger at a distance at distance center, and clocklike discrete speck distributes.According to contrast distance center place speck at a distance, can identify the two.Concrete control methods is as follows: at first make the two the center brightness normalization of tobacco transform domain image to be measured and standard tobacco and background changing area image, equate even two amplitude variations to be distinguished change the area image center brightness, make simultaneously that the ratio of each pixel and central pixel point brightness remains unchanged in the image separately; Be the center then with the central pixel point, the length of respectively getting certain pixel quantity up and down marks off a rectangle, and in the part in the statistical picture outside rectangle, brightness surpasses certain value G ₁The number N of pixel; Several standard tobacco leaves in the internal memory and background changing area image are repeated aforesaid operations, with the maximum N of the N of several transform domain images _MaxAs the upper limit, when the N of tobacco transform domain image to be measured greater than N _Max, then can regard as the image that includes foreign matter Calusena lansium band, otherwise then be tobacco leaf image.

4, identification signal transmission: after tobacco and foreign matter discerned, the signal of expression foreign matter position is transferred to the ECU of system.

5, reject foreign matter: ECU is controlled corresponding air nozzle with foreign body eliminating according to received foreign matter position signalling.

Claims (2)

1, the textural characteristics ONLINE RECOGNITION method that foreign matter detects in the article may further comprise the steps:

(1) image capture: on production line, with the article to be measured in the CCD line array video camera picked-up certain limit and the image of foreign matter mixture;

(2) image transmission: reach in calculator memory or the data signal embedded processing systems with the position signalling of image pick-up card with taking the photograph view data and article flow;

(3) foreign matter identification is handled: at first according to the colour information of image, utilize based on whether including foreign matter in the pixel colourity diagnostic method recognition unit; If can't utilize the colourity of pixel to judge, then carry out texture recognition according to the FFT conversion of unit;

(4) identification signal transmission: after article and foreign matter discerned, the signal of expression foreign matter position is transferred to ECU;

(5) reject foreign matter: ECU is controlled corresponding air nozzle with foreign body eliminating according to the foreign matter position signalling that is received;

It is characterized in that:

The foreign matter identification of step (3) has been adopted the recognition methods of carrying out texture analysis based on the FFT conversion in handling, and this method may further comprise the steps:

1. gray processing: the coloured image of standard article is transformed to gray level image;

2. smothing filtering: gray level image being carried out medium filtering, eliminate noise, promptly is use a sliding window on the gray level image of gained, the gray scale of each pixel in the window is sorted by size, with the former gray scale of its intermediate value replacement window center pixel;

3. FFT conversion: several standard images of items and background image are carried out the FFT conversion, the view data after the conversion is deposited in the internal memory; Treat detected image and carry out the FFT conversion, obtain transform domain image;

4. discern according to changing image: the FFT changing image of testing image and standard article and background is compared, concrete grammar is as follows: at first make the two center brightness normalization, equate even two amplitude variations to be distinguished change the area image center brightness, make simultaneously that the ratio of each pixel and central pixel point brightness remains unchanged in the image separately; Be the center then with the central pixel point, the length of respectively getting certain pixel quantity up and down marks off a rectangle, and in the part in the statistical picture outside rectangle, brightness surpasses certain value G ₁The number N of pixel; Several standard article in the internal memory and background image are repeated aforesaid operations, with the maximum N of the N of several transform domain images _MaxAs the upper limit, when the N of testing image greater than N _Max, then can regard as the image that includes foreign matter, otherwise then be images of items.

2, ONLINE RECOGNITION method according to claim 1, it is characterized in that: in step (1), use light emitting diode light source as a setting, with the color close look as a setting with the article color, and all make a video recording in the determinand top and bottom simultaneously, the image on determinand two sides is all discerned.

Images