Machine vision is the use of machines instead of the human eye to make measurements and judgments. The machine vision system refers to the conversion of the captured target into an image signal through a machine vision product (ie, an image pickup device, which is divided into CMOS and CCD), and is transmitted to a dedicated image processing system, which changes according to pixel distribution, brightness, color, and other information. Digital signals are generated; the image system performs various operations on these signals to extract the characteristics of the target, and then controls the on-site equipment actions according to the results of the discrimination.
Machine vision systems are characterized by increased flexibility and automation of production. In some dangerous working environments that are not suitable for manual work or artificial vision is difficult to meet the requirements, machine vision is often used instead of artificial vision; at the same time, in the process of large-scale industrial production, the use of artificial vision to check the product quality is inefficient and the precision is not high. Machine vision inspection methods can greatly increase production efficiency and automation of production. Moreover, machine vision is easy to implement information integration and is the basic technology for achieving computer integrated manufacturing.
A typical industrial machine vision application system includes the following parts: light source, lens, CCD camera, image processing unit (or image capture card), image processing software, monitor, communication/input/output unit, etc. First, the camera is used to obtain the image signal of the target to be measured, and then it is converted into a digital signal by A/D conversion to a dedicated image processing system. Based on the information such as pixel distribution, brightness, and color, various operations are performed to extract the characteristics of the target, and then Then, according to a preset criterion, the judgment result is output, and the drive actuator is controlled to perform corresponding processing. Machine vision is a comprehensive technology, including digital image processing technology, mechanical engineering technology, control technology, light source lighting technology, optical imaging technology, sensor technology, analog and digital video technology, computer software and hardware technology, human-machine interface technology. Machine vision emphasizes practicality, requires adaptability to the harsh environment of industrial sites, requires reasonable cost-effectiveness, common industrial interfaces, high fault tolerance and safety, and has strong versatility and portability. It emphasizes Real-time, requiring high speed and high precision.
The output of the vision system is not the image video signal, but the detection result after the operation processing, such as the size data. After the host computer such as PC and PLC obtains the test result in real time, the command motion system or I/O system performs corresponding control actions such as positioning and sorting. From the visual system operating environment classification, can be divided into PC-BASED system and PLC-BASED system. PC-based systems take advantage of its openness, high programming flexibility and a good Windows interface, while the overall system cost is lower. Take DATA TRANSLATION Inc. as an example. The system contains a high-performance image capture card. Generally, it can be used to access multiple lenses. There are several levels from low to high in supporting software, such as C/C++ programming in the Windows 95/98/NT environment. DLL, the visual control activeX provides the graphical programming environment under VB and VC++, even the object-oriented machine vision configuration software under Windows, users can use it to develop complex and advanced applications quickly. In PLC-based systems, vision acts more like an intelligent sensor. The image processing unit is independent of the system and exchanges data with the PLC via the serial bus and I/O. System hardware generally uses high-speed dedicated ASIC or embedded computer for image processing. The system software is solidified in the image processor. The menu displayed on the monitor is configured through a simple device similar to the game keyboard, or the software is developed on the PC. download. The PLC-based system embodies the characteristics of high reliability, integration, miniaturization, high speed, and low cost. Representative manufacturers are Panasonic of Japan and Siemens of Germany.
Siemens, Germany has more than 20 years of experience in industrial image processing. SIMATIC VIDEOMAT is the first high-performance monochrome and color image processing system and has become an extremely important product in the SIMATIC automation system. The SIMATIC VS710, introduced in 1999, is the industry's first intelligent, integrated, distributed gray-scale industrial vision system with PROFIBUS interface. It integrates an image processor, CCD, and I/O in a small package. Inside the chassis, provides PROFIBUS networking (12Mbps communication speed) or integrated I/O and RS232 interface. More importantly, through the configuration of Pro Vision parametric software under PC WINDOWS, VS 710 combines the flexibility of PC, PLC reliability, distributed network technology, and integrated design for the first time, making Siemens A perfect balance was found between the PC and PLC systems. Application of Machine Vision System in Printing and Packaging
1, automatic print quality inspection
The detection system used in the automatic print quality inspection equipment is to first take a standard image using a high-definition, high-speed camera lens, and set a certain standard on this basis; then, the image to be detected is taken and the two are compared. The CCD linear sensor converts the change in the light quantity of each pixel into an electronic signal. After the comparison, if the detected image is different from the standard image, the system considers the detected image to be a defective product. The various errors generated in the printing process are only different from the standard image and the detected image for the computer, such as defects such as smudges, ink spots, and the like.
The earliest used for print quality inspection is the technology of comparing the gray scale of the standard image with the detected image. Now the more advanced technology is based on the RGB three primary colors for comparison. Where is the difference between fully automated machine testing and human eye detection? When human eyes are regarded as examples, when we concentrate on a printed matter, if the contrast of the printed matter is strong, the smallest flaw that the human eye can find, Contrast color is not less than 0.3mm defects; but it is difficult to maintain a consistent and stable visual effects depending on human capabilities. In another case, if defects are found in prints of the same color, especially in a light color system, the defects that the human eye can detect need at least 20 gray levels. An automated machine can easily detect a 0.10mm defect, even if the defect is only one grayscale difference from the standard image.
However, in terms of practical use, even the same panchromatic contrast system has different ability to discriminate color differences. Some systems can find contours and defects with large variations in color, while others can identify very small defects. For white cardboard and some simple style prints, such as the Japanese KENT cigarette label, the United States Marlboro cigarette label, simple detection may be sufficient, and most of the domestic print, especially various labels, has many characteristics, With too many flash elements, such as gold, silver cardboard, hot stamping, embossing, or polishing prints, this requires that the quality inspection equipment must have sufficient ability to find the smallest grayscale differences, perhaps 5 shades of gray. The difference in level may be a stricter one gray level difference. This point is crucial to the domestic label market.
The accuracy of the comparison between the standard image and the printed matter being examined is a key issue in the detection equipment. Usually, the detection equipment collects the image through the lens. In the middle part of the scope, the image is very clear, but the image at the edge may be false. Shadow, and the detection result of the phantom part directly affects the accuracy of the entire test. From this point of view, if only the comparison of the full width area is not suitable for some fine prints. If the obtained image can be subdivided again, for example, the image is divided into 1024dpi X 4096dpi or 2048dpi X 4096dpi, the detection accuracy will be greatly improved, and at the same time the false detection of the edge portion is avoided, so that the detection result is more stable.
The use of inspection equipment for quality inspection provides real-time reporting of the entire process and detailed and complete analysis reports. The on-site operator can rely on the timely detection of the automatic detection equipment and timely adjust the problems in the work according to the real-time analysis report. It is possible that the reduction will be not only a percentage point scrap rate, the manager can analyze the report based on the test results. Tracking the production process is more conducive to the management of production technology. Because of the high-quality testing equipment required by customers, not only does it stop at the quality of the printed matter, but it also requires the ability to analyze afterwards. Some quality inspection equipment can not only improve the qualification rate of finished products, but also help manufacturers to improve the process and establish a quality management system to achieve a long-term stable quality standards.
2, gravure printing machine position control and product testing
The video images of the printed products are continuously taken by a camera set on the production line, and the speed of the recording is adjustable at 30 frames/s or less. The image captured by the camera is first quantified, the analog signal is converted into a digital signal, and a key frame that effectively represents the content of the lens is extracted from it and displayed on the display. For a frame of image, an analysis method for a still image may be used to process. Through size measurement and multi-spectral analysis, color patches on a video image may be identified, and color space and color parameters of the color patch may be obtained as well as some other correlations.
Due to various factors, various types of noise may occur, such as Gaussian noise, salt and pepper noise, and random noise. Noise brings many difficulties to image processing. It has a direct impact on image segmentation, feature extraction, and image recognition. Therefore, real-time acquired images need to be filtered. Image filtering requires the removal of noise outside the image while maintaining the details of the image. When the noise is Gaussian noise, the most commonly used linear filter is easy to analyze and implement; however, the linear filter has poor filtering effect on salt and pepper noise, and the traditional median filter can reduce the salt and pepper noise in the image, but the effect Not ideal, that is, fully dispersed noise is removed, and the noise close to each other will be retained, so when the salt and pepper noise is more serious, its filtering effect is significantly worse. This system improved median filtering method. The method firstly obtains the median value after removing the maximum and minimum grayscale pixels in the noise image window, and then calculates the difference between the median value and the corresponding pixel grayscale value, and then compares it with the threshold to determine whether to use the obtained value. Instead of the grayscale value of the pixel.
Image segmentation detects color patches and separates them from the background at this stage. The edge of the object is represented by gray-scale discontinuity. The edge type can be divided into two types. The first is the step edge, and the pixels on both sides are The gray value is significantly different; the second is the roof-like edge, which lies in the change of the gray value from increasing to decreasing turning point L. For the step edge, the second-order directional derivative is zero-crossing at the edge, so the differential calculation is available. Son to do edge detection operator. The differential operator class edge detection method is similar to the high-pass filter in the high spatial domain and has the effect of increasing high-frequency components. Such operators are quite sensitive to noise. For step edges, the commonly available operators include the gradient operator Sobel operator. Child and Kirsh operators. Laplace transforms and Kirsh operators are available for roof-like edges. Because the color mark is rectangular and the gray levels of adjacent edges are quite different, edge detection is used to divide the image. Here we use Sobert edge to perform edge detection. It uses the local difference operator to find the edge, which can separate the color marks better. In the actual detection process, a color image edge detection method is adopted, and an appropriate color base (such as intensity, chroma, saturation, etc.) is selected for detection. According to the type characteristics of the printing press, that is, the color of the printing machine and the characteristics of the layout, multi-threshold processing is performed to obtain binary maps of various colors.
The segmented image is measured and the object is identified by the measured value. Since the color patch is a rectangle with a regular shape, the following features can be extracted: (1) Calculate the area of ​​a rectangle by pixels, (2) Rectangularity, (3) ) Chromaticity (H) and Saturation (S), and then the spacing between the color patches is obtained according to the number of pixels in the interval between each color standard. Compared with the set value, the difference between the two is obtained, and a total of m measurements are performed to obtain the average value. The difference provides a corresponding adjustment signal to the digital AC servo adjustment section. To adjust the relative position of the color roller to eliminate or reduce misalignment. In feature extraction, multispectral image analysis is performed on the image to quantitatively represent color patches, such as the color of pixels in a color number image. The HIS format is used to obtain two parameters of the color information of each color standard: chroma and saturation. To check the quality of the ink. The statistical analysis of the binary image of each color and the matching of the template with the standard graphics are performed to measure the parameters such as the ink scraps during the printing process.
The printing machine is unwound from the uncoiler and passes through the printing units in sequence to carry out printing and drying of various colors. The winding machine performs the winding and each color printing prints a color mark on the edge of the printing material for color registration. The color markings are 10 mm horizontal and 1 mm wide. The marking lines of each adjacent color should be parallel to each other when they are accurately printed, and the vertical (vertical) distance is 20 mm. Video images of printed products are continuously taken by the cameras installed on the production line. Through the size measurement and multi-spectral analysis can identify the color image on the video, get the color space and the color parameters of the color L If the interval between adjacent two color color is greater than or less than 20 mm, it indicates overprinting deviation. The deviation signal is sent to the servo variable frequency drive unit to drive the AC servo motor to move the corresponding chroma correction roller ML up and down to extend or shorten the dynamic correction of the printing material from the previous unit printing plate roller to the printing plate roller.
3, application in the modern packaging industry
In the automated production of modern packaging industry, it involves a variety of inspections and measurements, such as the printing quality inspection of beverage bottle caps, bar codes and character recognition on product packaging, etc. The common feature of this type of application is the continuous mass production and the very high demands on the appearance quality. Normally this kind of highly repetitive and intelligent work can only be accomplished by manual inspection. We often see hundreds or even thousands of inspection workers behind the modernization lines of some factories to perform this process. While the factory has increased enormous labor costs and management costs, it still cannot guarantee a 100% inspection pass rate (ie, “zero defectsâ€), and today’s competition among enterprises has not allowed even 0.1% of defects to exist. In some cases, such as precise and rapid measurement of a small size, shape matching, color identification, etc., it is impossible for the human eye to perform continuously and stably, and other physical quantity sensors are also difficult to use. At this time, people began to consider the rapidity, reliability, and reproducibility of the results of the computer, thereby introducing robot vision technology.
In general terms, first the CCD camera is used to convert the captured target into an image signal, which is then sent to a dedicated image processing system and converted into a digital signal based on pixel distribution, brightness, and color information. The image system performs various operations on these signals. Extract the characteristics of the target, such as: area, length, quantity, position, etc. Finally, output the result according to the preset tolerance and other conditions, such as: size, angle, offset, number, pass/fail, yes/ Nothing. Machine vision is characterized by automation, objectivity, non-contact, and high precision. Compared with image processing systems in the general sense, machine vision emphasizes accuracy and speed, and reliability in industrial field environments. Machine vision is extremely suitable for measurement, inspection, and identification in mass production processes, such as the identification of printed characters on the surface of ICs, the identification of production dates on food packaging, and the inspection of label placement locations.
A typical vision system generally includes the following parts: a light source, a lens, a CCD camera, an image processing unit (or image capture card), an image processing software, a monitor, a communication/input/output unit, and the like. The output of the vision system is not the image video signal, but the detection result after the operation processing, such as the size data. After the host computer such as PC and PLC obtains the test result in real time, the command motion system or I/O system performs corresponding control actions such as positioning and sorting. From the visual system operating environment classification, can be divided into PC-BASED system and PLC-BASED system. PC-based systems take advantage of its openness, high programming flexibility and a good Windows interface, while the overall system cost is lower. Take DATA TRANSLATION Inc. as an example. The system contains a high-performance image capture card. Generally, it can be used to access multiple lenses. There are several levels from low to high in supporting software, such as C/C++ programming in the Windows 95/98/NT environment. DLL, the visual control activeX provides the graphical programming environment under VB and VC++, even the object-oriented machine vision configuration software under Windows, users can use it to develop complex and advanced applications quickly. In PLC-based systems, vision acts more like an intelligent sensor. The image processing unit is independent of the system and exchanges data with the PLC via the serial bus and I/O. System hardware generally uses high-speed dedicated ASIC or embedded computer for image processing. The system software is solidified in the image processor. The menu displayed on the monitor is configured through a simple device similar to the game keyboard, or the software is developed on the PC. download. The PLC-based system embodies the characteristics of high reliability, integration, miniaturization, high speed, and low cost. Representative manufacturers are Panasonic of Japan and Siemens of Germany.
4 Technological advances in machine vision
In machine vision systems, key technologies include light source lighting technology, optical lenses, cameras, image capture cards, image processing cards, and fast and accurate actuators. In machine vision applications, good light sources and lighting schemes are often the key to the success or failure of the entire system; they play a very important role; they are not simply illuminating objects. The cooperation of the light source and the lighting scheme should highlight the object features as much as possible; there should be obvious differences between the parts of the object that need to be detected and those that are not important; the contrast should be increased; and the overall brightness should be ensured; Changes in the position of the object should not affect the quality of the image. Transmitted light and reflected light are generally used in machine vision applications. For the case of reflected light, the relative positions of the light source and the optical lens, the texture of the surface of the object, and the geometrical shape and background of the object should be fully considered. The choice of light source must meet the required geometry, illumination brightness, uniformity, spectral characteristics of light emission, etc.; at the same time, the luminous efficiency and service life of the light source must also be considered. The optical lens is equivalent to the lens of the human eye; it is very important in the machine vision system. The quality of the imaging of a lens is good or bad; that is, whether it is good or not for aberration correction; it can be measured by the size of aberrations; common aberrations are spherical aberration, coma aberration, astigmatism, field curvature, distortion, color aberration, etc. .
The camera and the image acquisition card jointly complete the collection and digitization of the material images. High-quality image information is the original basis for the correct judgment and decision of the system; it is another key to the success of the entire system. Currently in machine vision systems, CCD cameras are widely used due to their compact size, reliable performance, and high definition. The CCD camera can be divided into two major categories of line array and area array according to the CCD device it uses. The linear CCD camera can only obtain one line of information at a time; the object being photographed must be moved in a straight line from the camera to obtain a complete image; therefore, it is very suitable for image detection of a material stream moving at a constant speed and at a uniform speed; The area array CCD camera can obtain the entire image information once. The processing of image signals is the core of the machine vision system; it is equivalent to the human brain. How to process and calculate the image; that is, the algorithm is embodied here; it is the key and difficult point in the development of machine vision system. With the rapid development of computer technology, microelectronics technology and large-scale integrated circuit technology, in order to improve the real-time performance of the system, many tasks for image processing can be accomplished with hardware; such as DSPs, dedicated image signal processing cards, etc.; Complete the parts of the algorithm that are very complex and not mature enough to be explored and changed.
From the point of view of the product itself, machine vision will increasingly rely on PC technology and will be more integrated with other controls and measurements such as data acquisition. Embedded products will gradually replace board products. This is a growing trend. The main reason is that with the rapid development of computer technology and microelectronics technology, the application of embedded systems has become more and more extensive, and in particular, the characteristics of low-power technologies have attracted people's attention. In addition, most of the embedded operating systems are based on C language. Therefore, the use of C high-level language for embedded system development is a basic task. The advantage of using high-level languages ​​is that it can improve work efficiency and shorten development. The cycle, more importantly, is that the developed products have high reliability, good maintainability, convenience for continuous improvement and upgrading. Therefore, embedded products will replace board-based products.
Because machine vision is a part of automation, machine vision is not without automation. Machine vision hardware and software products are gradually becoming the core system in different stages of the collaborative manufacturing process. Both users and hardware suppliers use machine vision products as production lines. The tools for information collection require a large number of "standardization technologies" for machine vision products. Intuitively speaking, it is necessary to gradually open up with the opening of automation, and secondary development can be carried out according to the needs of users. Today, automation companies are advocating an integrated software and hardware solution. In the next five to six years, machine vision manufacturers should not only be merely suppliers of products, but should gradually move toward system integrators of integrated solutions.
In the next few years, with the development of processing and manufacturing in China, the demand for machine vision will gradually increase. With the increase in machine vision products and the improvement of technology, the application of domestic machine vision will shift from the low end of the initial stage to the high end. Due to the involvement of machine vision, automation will develop in a smarter and faster direction.