A laser imaging method for machine vision detection of white contaminants in cotton

Abstract

The existing machine vision systems cannot efficiently detect white contaminants in cotton under the illumination of visible lights, because their color is the same or very close. To solve the problem, this article proposes an imaging method based on line lasers. Under the illumination of a line laser, the white contaminants and cotton showed the differences in the optical characteristic of their surface. Then, according to the features of the intensity of their reflected lights or the distribution of the fluff around their surfaces in the images, an example algorithm for identification of white contaminants from cotton was suggested. The experimental results indicated that, using our method, the mean successful detection rate of the typical white contaminants in cotton was over 87%.

Keywords

machine vision cotton white contaminants line laser

Contaminants, such as polypropylene baler twine, plastic films, packaging materials, ropes, small pieces of cloth and nylon, are almost invariably present in cotton during the cotton picking, storing, drying and transporting. Their presence in textile processing may cause quality problems and thus reduce the end-product’s value, as their physical and chemical properties are different from cotton, even the weight of contaminants in cotton is less than that of one millionth of the cotton.¹

Since the 1990 s, machine vision systems have been used in the textile industry for the removal of contaminants in cotton, according to their color or grey level differences under the illumination of visible lights^2–4 in the blow room. At this stage, the cotton has been sufficiently opened and the contaminants can be “seen” on the surface of the cotton tufts.^5,6 But so far, the systems cannot efficiently distinguish most of the white contaminants from cotton because they are the same or close in color.^7,8 Because the grayscale histograms of these images were of a single peak, most of the suggested algorithms for image processing were too complex to be used on line, or some of them only for a special case of the inspections.^9–13 As the result, manual sorting methods of white contaminants are still used in textile processing, which is time-consuming, inefficient and highly costly.¹⁴

To solve the problems, ultraviolet lights were used for the detection of white contaminants,^15–17 but a large proportion of the white contaminants is non-fluorescent, which still cannot be sorted out from cotton. Dogan et al. and Pai et al. researched the 2D transmission image acquisition using X-ray for thorough analysis of the contaminants.^18,19 However, because of the safety problem and complex image processing, this system based on X-ray still needs further improvement for real-time applications.

In recent years, research on image acquisition based on infrared bands became a hotspot. Church et al. found the difference of the spectrum peaks of cotton and contaminants in infrared bands, and made a suggestion of using the lights in the range of 2250–2400 nm spectrum.²⁰ Himmelsbach et al. tried to setup a spectral database by ATR/FT-IR analysis to identify the typically classified foreign matters in cotton.²¹ Guo et al. and Jia et al. researched the identification of contaminants in lint cotton based on infrared lights and concluded that some of the white contaminants can be identified in infrared bands.^22,23 But the speed of spectral analysis is slow and the equipment cost is high. Böhmer et al. developed a camera system based on two channels near infrared camera for detecting some white or transparent foils in raw cotton.²⁴ Yang et al. presented the best infrared bands for inspecting contaminants.²⁵ Their research pointed out the possibility of using infrared lights for the detection of white contaminants. However these applications are limited until advanced signal processors and infrared cameras of high speed and high resolution are produced in a large scale.

Recently, the line laser was widely used in the 3D measurement field on account of its advantages of good directionality, high energy, monochromaticity and coherence. Hua et al. found that typical white contaminants, such as nylon belts, acrylic belts, nylon cords, different types of white synthetic fibers and plastic films, and so on, reflected a strong light in a special angle on the cotton surface, which was different from cotton under the illumination of a line laser.²⁶ The difficulty is that in the machine vision inspection the cotton surface is uneven so that the system cannot exactly capture the reflected light from different angles.

This article proposed a new method for distinguishing typical white contaminants from cotton using line laser imaging. The experiment showed that by acquisition of the images of the microscopic characteristics on the surface of the white contaminants and cotton, most of the white contaminants can be detected.

Laser imaging system

The imaging system of this article is shown in Figure 1.

Figure 1.

Imaging system.

Here, a light of line laser was used to “cut” a cotton layer at a certain angle. The upper part of the cutting cross-section of the sample was penetrated and brightened, where the fibers absorbed and reflected a part of light energy, as shown in Figure 2.

Figure 2.

Laser cutting cross-section of the sample.

Cotton fibers are slender and curl about 20 um in diameter, up to 45 mm in length. In the natural state, they are loosely wound and wrapped together. The appearance of a cotton mass looks like a cloud or mist. From the microstructure point of view, fluff and free filament fibers are always distributed on the surface of cotton. While, most of white contaminants, such as white plastic mulches, white packing paper and strings, white cloths and hemp ropes, whose surfaces are compact and smooth, in the shape of a strip, sheet or wire with a clear edge, but without any fluff or filaments on the surface.

In the laser cutting cross-section of the samples, the white contaminants and the compact cotton mass are bright and white, shown as a high light block (HLB), and more important, around the cotton, there are always lots of high light sparks (HLS) which come from the cotton fluff and free filaments, but nearly nothing around the white contaminants, as shown in Figure 3. According to our statistical analysis, the white contaminants can be distinguished from cotton by calculation of the HLB and HLS.

Figure 3.

Contaminants with cotton in a laser cutting cross-section.

The example images of different types of white contaminants in cotton are shown in Figures 4(a)–(h), where the images in the left column, in the center column and in the right column were respectively captured under the illumination of white LED, white LED plus line laser, and line laser only.

Figure 4.

Sample images of cotton with contaminants under illumination of LED, LED + laser or only laser. (a) White paper, (b) white plastic cardboard, (c) white plastic cord, (d) polyethylene foamed sheet, (e) semitransparent plastic mulch, (f) white feather, (g) white nylon cord, (h) semitransparent polypropylene bag.

Example algorithm for image processing

The flow chart of an algorithm for image processing is shown in Figure 5.

Figure 5.

Flow chart of an algorithm for image processing.

An image acquired under the laser illumination can be divided into two parts: One part is loose cotton with the background in lower pixel values; and another in higher pixel values, as shown in Figure 6(a).

Figure 6.

(a) Original image, (b) histogram, (c) binary image, (d) HLB image, (e) dilated HLB area, (f) HLS in dilated HLB area.

Firstly, the loose cotton with the background is easily removed by employing a simple thresholding method, as the histograms of the grayscale images are of double peaks, as shown in Figure 6(b). Theoretically speaking, any value at the valley bottom can be taken for the binarization. In the experiment the thresholding value for all the images was taken at 180. The binary image is shown in Figure 6(c).

Secondly, by the size of a single object (connected component) in the binary image, HLB (compact cotton and contaminants) can be separated from HLS (cotton fluff and free filaments), as shown in Figures 6(d) and (f). In the experiment, from statistical analysis, the minimum size of any HLB was 25 pixels (T_b), and maximum size of any HLS was 3 pixels (T_s).

Next, in order to distinguish contaminants from the compact cotton (both are HLB), a dilation operation²⁷ was used for each HLB. Mathematically, the dilation of X by B, denoted X ⊕ B, is expressed as:

X \oplus B = {z | (\overset{\land}{B}) z \cap X \neq φ}

where φ is the empty set and B is the structuring element. In the experiment, each of the HLBs was dilated by a disk-shaped structuring element with a radius of 14 pixels. The expanded HLB is shown in Figure 6(e).

Then, the density of HLS in the ring region around a HLB was calculated, as shown in Figure 6(f), which was defined as a proportion of the sum of HLS to the perimeter of the dilated HLB in the ring region. If the density of the HLS is higher than the threshold T₀, the HLB is for cotton; otherwise, it is for a contaminant. In this experiment, the threshold T₀ was 0.086, based on statistical analysis of our sample set.

Experiment results and discussion

The experimental system of this article is shown in Figure 1, above. The images were captured by a Basler A602fc camera, resolution in 640 × 480 pixels. A line laser at 50 mw of power, wavelength 650 nm, was selected as the illumination source. As a test sample, a continuous opening lint layer formed so as to be 50 cm wide and ∼2–3 cm high was put on a conveyor surface. The camera optical axis was perpendicular to the light plane formed by the line laser.

12 types of typical white contaminants, 10 pieces of each type, were selected for the experiment, as shown in Figure 7. In the experiment, 10 images of each sample in different background of cotton were taken. In total, 1200 images for cotton with contaminants, and 1200 images for pure cotton were acquired for the test.

Figure 7.

Foreign fiber samples. (1) White paper, (2) white plastic film, (3) semitransparent plastic mulch, (4) white cloth, (5) white density foam plastic, (6) white nylon cord, (7) cotton string, (8) white plastic cord, (9) white plastic cardboard, (10) semitransparent polypropylene bag, (11) polyethylene foamed sheet, (12) white feather.

The algorithm was implemented and validated by MATLAB, using a computer with Intel(R) Core™ i5-2540 M@2.60 GHz CPU and 2 GB RAM operating a Windows 7 system.

The experimental results are shown in Table 1.

Table 1.

Results of detecting contaminants

Number	Contaminants	Total images	Successfully detected images
1	White paper	100	98
2	White plastic film	100	97
3	Semitransparent plastic mulch	100	95
4	White cloth	100	92
5	White density foam plastic	100	93
6	White nylon cord	100	90
7	Cotton string	100	78
8	White plastic cord	100	94
9	White plastic cardboard	100	83
10	Semitransparent polypropylene bag	100	76
11	Polyethylene foamed sheet	100	47
12	White feather	100	35
13	Pure cotton	1200	1119

If A is the number of the successful detected images and D is the number of the total images, the detection rate is calculated as follows:

Detection_Rate = \frac{A}{D} \times 100 %

Where, the successful detection rate for all the images was 87.4%. In detail, for the contaminants with smooth and compact surfaces, such as the white paper, white plastic films, semitransparent plastic mulch, white plastic cords, white density foam plastic, white cloths and white nylon cords, the average successful detection rate was 94%. While for the contaminants with comose and hairy surfaces, including the white feather and polyethylene foamed sheets, the successful detected rate was only 41%. The main reason is that the white contaminants were distinguished from cotton by the distribution and density of the flosses and free filaments, but not by the color.

Basically, there were two error sources in the experiment. One was from the imaging method, by which a few of contaminants, such as a feather or polyethylene foamed sheet, could not be distinguished from cotton because their surfaces are comose or hairy, very similar to cotton; while similarly a few of the cotton masses without fluff on the surface, because the fibers were stuck together, could not be distinguished from the contaminants. The other error source was the algorithm, which could not identify the HLB if a contaminant connected with a compact cotton layer and they formed a single HLB.

Conclusions

The existing imaging methods in machine vision systems cannot efficiently distinguish non-fluorescent white contaminants from cotton because they are the same or close color under the illumination of the visible lights and ultraviolet lights. A laser imaging method for detecting the white contaminants on the surface of cotton is presented in the paper, based on the fact that in the image of the laser cutting section, there are lots of HLSs around the cotton outside layer, which is different from the white contaminants. The experiment of this article indicated that, according to their difference, the mean successful distinguishing rate of the typical white contaminants from cotton was over 87% by the suggested algorithm.

In the experiment, the detectable minimum size of a contaminant was 25 pixels, about 1.4 mm², and the movement speed of the imaging system was 18 m/min. For engineering applications, according to the laser imaging method, the imaging speed and accuracy can be improved by more powerful laser and fast cameras, and the detectable minimum size of a contaminant can be changed by employing a different size of HLB in the algorithm, or a more powerful algorithm.

Line lasers are of low cost, small in size and easy to use. Compared with infrared and X-ray systems, the laser imaging system is simple and easily integrated into an existing sorting system.

Next, research work will focus on engineering applications for the method. Different types of line laser at different wavelengths and powers will be used for the image acquisition of white contaminants with different types of cotton in different textile processing stages. And, besides the optimization of the laser imaging method and the algorithm, as we have mentioned before,²⁸ based on human visual attention, a group of programmable area cameras will be used for the image acquisition and suspicious image location, and then, after the removal of the most redundant image data, a host computer will be used for sorting out contaminants from cotton.

Footnotes

Funding

This work was supported by the National Natural Science Foundation of China (grant number NSFC31371536).

References

International Textile Manufacturers Federation (ITMF). Cotton Contamination Surveys (2001–2003–2005–2007–2009–2011). Zurich: International Textile Manufacturers Federation, 2012, on CD ROM.

Lieberman

Bragg

Brennan

. Determining gravimetric bark content in cotton with machine vision. Text Res J 1998; 68(2): 94–104.

Tian

Gao

. A machine vision system for on-line removal of contaminants in wool. Mechatronics 2006; 16(5): 243–247.

Yang

Zhu

. A new approach for image processing in foreign fiber detection. Comput Electron Agr 2009; 68(1): 68–77.

Ding

. A fast feature extraction algorithm for detection of foreign fiber in lint cotton within a complex background. Acta Automatica Sinica 2010; 36(6): 785–790.

Tantaswadi

Vilainatre

Tamaree

. Machine vision for automated visual inspection of cotton quality in textile industries using color isodiscrimination contour. Comput Industr Eng 1999; 37(1): 347–350.

Finlayson

Hordley

Schaefer

. Illuminant and device invariant colour using histogram equalization. Pattern Recogn 2005; 38(2): 179–190.

Zhang

Frittsb

Goldmana

. Image segmentation evaluation: A survey of unsupervised methods. Comput Vis Image Und 2008; 110(2): 260–280.

Kang

Kim

. Objective evaluation of the trash and color of raw cotton by image processing and neural network. Text Res J 2002; 72(9): 776–782.

10.

Ding

Huang

. An automated cotton contamination detection system based on co-occurrence matrix contrast information. Intel Comput Intell Sys 2009; 4: 517–521.

11.

Feng

Zhang

Yang

. Foreign fiber recognition and detection algorithm based on RGB color space. J Shandong Univ (Eng Sci) 2009; 5: 014–014.

12.

Zhang C, Feng X, Li L, et al. Detection of foreign fibers in cotton on the basis of wavelet. In: Proceedings of the 2nd international conference on signal processing systems (ICSPS), Dalian, China, 5–7 July 2010, Vol. 2, pp.304–308.

13.

Mehta

Kumar

. Detection of foreign fibers and cotton contaminants by using intensity and Hue properties. Int J Adv Electron Engin 2010; 1(1): 230–240.

14.

Yang

Zhu

. A new approach for image processing in foreign fiber detection. Comput Electron Agr 2009; 68(1): 68–77.

15.

Hogan

. Spectroscopy focus: Tracking cotton trash via fluorescence. Photonics Spectra 2007; 41(8): 98–99.

16.

Luo

Zhu

. Fluorescent effect of ultra violet and its application in detection of foreign fibers in cotton. J Henan Univ (Nat Sci) 2007; 2: 63–66.

17.

Gamble

Foulk

. Quantitative analysis of cotton (Gossypium hirsutum) lint trash by fluorescence spectroscopy. J Agr Food Chem 2007; 55(13): 4940–4943.

18.

Dogan M, Sari-Sarraf H and Hequet E. Cotton trash assessment in radiographic x-ray images with scale-space filtering and stereo analysis. In: Machine vision applications in industrial inspection XIII, San Jose, USA, Proceedings of the society of photo-optical instrumentation engineers (SPIE), 17–18 January 2005, Vol. 5679, pp.276–287.

19.

Pai

Sari-Sarraf

Hequet

. Recognition of cotton contaminants via X-ray microtomographic image analysis. IEEE Ind Appl Mag 2004; 40(1): 77–85.

20.

Church

O'Neill

Woodhead

. Detection of fibrillated polymeric contaminants in wool and cotton yarns. Appl Spectrosc 1998; 56(8): 1039–1046.

21.

Himmelsbach

Hellgeth

McAlister

. Development and use of an attenuated total reflectance/Fourier transform infrared (ATR/FT-IR) spectral database to identify foreign matter in cotton. J Agri Food Chem 2006; 54(20): 7405–7412.

22.

Guo

Ying

. Detection of foreign materials on the surface of ginned cotton by hyper-spectral imaging. Trans Chin Soc Agri Engin 2012; 28(21): 126–134.

23.

Jia

Ding

. Novel method of detecting foreign fibers in lint by fiber’s infrared absorption characteristic. J Infrared Millim W 2005; 24(2): 147–150.

24.

Böhmer S, Budzier H, Krause V, et al. Two-channel NIR camera system to detect foreign matter in raw cotton. In: Proceedings of 8th Quantitative InfraRed Thermography (QIRT), Padova, Italy, 27–30 June 2006, pp.53–59.

25.

Yang

. Selection of optimal band for detecting foreign fibers in lint cotton using spectroscopic analysis. Trans Chin Soc Agr Engin 2009; 25(10): 186–192.

26.

Hua

Qiao

. White foreign fibers detection in cotton using linear laser. Trans Chin Soc Agr Mach 2012; 43(2): 181–185.

27.

Boomgaard

Balen

. Methods for fast morphological image transforms using bitmapped binary images. CVGIP—Graph Model Im 1992; 54(3): 252–258.

28.

Shi

. A new machine vision system with visual attention function. J Inform Comput Sci 2010; 7(14): 3011–3018.