Machine vision is to simulate human vision features from your computer to allow the machine to obtain the relevant visual information and be understood. Principle be divided into two parts "as" and "sleep", "view" is the display information to the outside into a digital signal back to the computer, hardware need to rely on a set of solutions, including a light source, a camera, an image acquired by the imaging card, visual sensors; "feel" is the computer digital signal processing and analysis, mainly software algorithms.
Machine vision applications in a wide field of industry, core functions comprising: measuring, detection, identification, location and the like . Chain market can be divided into an upstream stage member, midstream system integration / machine equipment market and downstream applications. Machine Vision upstream light source, lens, industrial camera, frame grabber, the image processing hardware and software providers software , integrated midstream and machine equipment providers, wider downstream applications, mainly downstream market include electronics manufacturing industry, automobile, printing and packaging, tobacco, agriculture, medicine, textiles, and transportation and other fields.
Machine Vision global markets mainly in North America, Europe, Japan and China, according to statistics, in 2014, the global machine vision system and component market size was $ 3.67 billion, in 2015 the global machine vision system and component market size is 4.2 billion US $ 2016 machine vision systems and components for the global market size of $ 6.2 billion, 2002--2016 years of market CAGR of around 12%. The machine vision system integration, according to the North American market data to estimate, about 6 times vision systems and components market.
China started in machine vision technology introduction in the 1980s, along with the entire line of 98-year semiconductor plant introduction, but also into machine vision systems, machine vision 06 years ago domestic product focused on foreign manufacturers, modest in size, 2006 start, industrial machine vision applications customer base began to expand into the field of detection printing, food, etc., in 2011 the market began rapid growth , with the escalation of increased demand and manufacturing labor costs, coupled with the rapid development of computer vision technology, increasingly the more machine vision solutions penetrated into all areas of machine vision in 2016 the market size has reached nearly 7 billion yuan .
Machine vision, defect detection, the application is one of the most machine vision functions, various kinds of information detected main surface of the product. In modern industrial automation production, the continuous mass production in each process has a certain defect rate, although the rate alone look small, but multiplied has become difficult for enterprises to improve the yield of the bottleneck, and after the complete process then remove defective costs would be much higher (e.g., if there is a positioning deviation of the solder paste printing process, and the problem until after the test line was discovered chip placement, repair cost would be less than 100 times the original cost), Therefore, timely detect and reject defective quality control and cost control is very important, and it is an important cornerstone of the manufacturing sector further escalation.
In the testing industry, compared with human vision, machine vision obvious advantages
1) high accuracy : human vision is 64 gray scale, and resolution of the minute weak target; machine vision can significantly improve gray scale, while observing a target micron;
2) speed : humans can not see the target is fast moving, the machine shutter time is up microsecond;
3) high stability : a human machine vision to solve the very serious problems of instability, MVI is very boring and hard labor in the industry, no matter how you design a reward system, relatively high missing rate will occur. But the machine vision inspection equipment is no fatigue, no mood swings, as long as you write something in the algorithm, each will seriously implement. In quality control results in greatly enhanced controllability.
4) the integration of information and retention : the amount of information obtained is a comprehensive machine vision and traceability, information can be easily integrated and retained.
2
In recent years, the rapid development of machine vision technology
1) image capture technology is developing rapidly
CCD, CMOS and other more mature firmware image sensing device dimensions shrink, the number of cells and data rates continue to improve, enhance the resolution and frame rate speed can be said that with each passing day, more and more rich product line, gain, shutter and constantly optimize the signal to noise ratio and other parameters, to a comprehensive selection of the light source, lens and camera through a core test indicators (the MTF, distortion, signal to noise ratio, the light source brightness, uniformity, color temperature, the image forming system Synthetic evaluation etc.), such that many difficult issues before the image is constantly breaking.
2) The image processing and pattern recognition developed rapidly
Image processing, the edge information of an image with high precision, low-contrast defects had much background noise mixed in the detection of the start is difficult to directly obtain resolution.
The pattern recognition process itself can be regarded as a marker, on the basis of a certain measure or observed on the pattern to be identified is divided into the respective modes to. Image recognition using a lot about the main decision-making theory and structural approaches. Basic theory decision is a decision function, using its pattern vector classification, described is the timing (e.g., statistical texture) basis; core structure is the method of the object is decomposed into primitive mode or modes, and different object structure different cell string (or character string), through a predetermined pattern of an unknown object using the primitive to determine a boundary encoding, to give the string, then the string is determined according to its genus. On feature generation, many new emerging algorithms, including wavelet, wavelet packet, fractal feature, and a unique two component based on the analysis; but also on the sub-support vector machine, deformable template matching, linear and non-linear classifier design, and so on continue to extend.
3) the depth of learning to bring breakthrough
Traditional machine learning on feature extraction and mainly rely on people to create a logical analysis, and the depth of learning is characterized by learning simple analog MLP workings of the brain, the neural network constructs a depth (e.g., a convolutional neural network), build complex wherein learn mapping and output, the training process at all levels will be continuously optimized. In specific applications, such as automatic ROI segmentation region; punctuation positioning (flexible unknown detecting defects by visual Simulator); severely noisy image can not be re-described detecting defects such as orange peel or quantify flaw; resolution detection of true cover glass false defects and so on. As more and more based on the depth learning machine vision software market (including Switzerland vidi, South Korea SUALAB, Hong Kong's ASTRI, etc.), the depth of learning to enabling machine vision will become increasingly apparent.
4) 3D visual development
3D vision still in its infancy, in many applications the use of 3D surface reconstruction, including navigation, industrial inspection, reverse engineering, mapping, object recognition, classification and measurement, etc., but the accuracy problems limit the use of many visual 3D scene, the current application engineering is the first to roll out the logistics in the standard volume measurement, I believe that this huge future potential.
3
Shipping to replace the manual visual inspection, machine vision there are still many difficulties to be broken:
1) The light source and imaging: Imaging machine vision quality is the first step, the surface due to the different reflective materials, refractive extraction problems will affect the characteristics of the object under test, and therefore can be said that the light source and the imaging machine vision to capture the first a storm. Such as now glass, reflective surface scratch detection, etc., often problems are stuck on different integrated imaging defects .
2) heavy noise in low-contrast image feature extraction: under heavy noise environments many cases, difficult to true and false identification of defects , which is the reason many scenes certain false detection rate is always there, but this edge and by the imaging characteristics extraction of rapid development, has constantly made breakthroughs that.
3) identification of unexpected defects : in the application, is often given some specific defect pattern, using machine vision to identify them in the end has not occurred. But often the case is that many obvious flaws, because before did not happen, or happens too diverse patterns, are missed. If people do change, though they do not allow him to detect this defect in the operation process documents, but he will notice, in order to have a greater chance to catch it, and machine vision at this point the "wisdom" is still difficult breakthrough.
4
The machine vision industry chain case
1) upstream market stage member
Includes a light source, lens, industrial cameras, frame grabbers, image processing software provider, in recent years, smart cameras, industrial cameras, lights, and the board has maintained a growth rate of not less than 20%. According to China Machine Vision Industry Alliance (CMVU) survey, China has now entered the international brand of nearly more than 200 machine vision (such as Cognex, Da Ersa, Baumer and other core components manufacturers represented to Keyence , Omron, Panasonic, Bonner, NI as the representative of the core to get involved in machine vision components and systems integration), China's own brand of machine vision also has more than 100 (such as Kang, Hua Rui UNITA extension optoelectronics, Divine vision, Shenzhen Rui Chan, Shanghai Fang Cheng, the Hyperion create electrical, etc.), machine vision agents more than 300 kinds of products (such as Shenzhen, Hong Fu vision, Microview, Sambo Industrial, Lingyun, sunlight vision, etc.). Many domestic markets machine vision components are agents of foreign brands from the beginning, many companies have better cooperation with foreign counterparts, and such cooperation has a certain exclusivity, which gives potential entrants bring a certain threshold, and therefore high-quality products agents also have good market competitiveness and profit performance . At the same time, to Kang, Hua Rui as the representative of the core components of domestic industrial vision is rapidly rising.
2) midstream system integrators and machine equipment market
System integration and domestic midstream machine equipment manufacturers more than 100, they can provide an integrated machine vision solutions to industry automation companies such as Lingyun, Microview, Jia Heng, Adlink, sunshine vision, tripod letter, Daheng images. As domestic and international products is still not a small gap, many of the middle reaches of the whole system integrators and equipment manufacturers is still doing the core components of the trade from them, so much the choice of visual products, is still more popular foreign brands. Domestic brands to promote their hardware and software products, often need to develop its own program integration capabilities in order to better face market competition.
3) downstream application markets
Downstream machine vision, mainly offer non-standard automation integrated solutions to the end-user companies, property industry is very strong, the core competitiveness is a comprehensive understanding of the industry and the production of many types of technology and integration. Due to the change of the automation industry has a certain periodicity, by the industry as a whole to upgrade the speed, volume and profit impact of the situation, so look at the past two years, boosting the popularity of machine vision applications, the most important thing in electronics manufacturing, followed by automobiles and pharmaceuticals .
i semiconductor and electronics manufacturing industry: from the domestic application profile view of the machine vision industry, 46% are concentrated in the electronics and semiconductor manufacturing industries, including the classification cut wafer manufacturing, PCB detection (backsheet, the inner / outer plates , finished look final inspection, etc.), detecting the SMT placement, the whole process of the LCD AOI detected defects, surface defect detection of various components 3c, 3c appearance testing
ii car: measurement error and geometry, surface and internal body assembly detected, defect detection part, a gap detection
iii printing, packaging, testing: Tobacco housing printing, plastic printing plates and food packaging, pharmaceutical packaging and printing
iv agriculture: agricultural products on grading, inspection and classification
Textile v: for different fiber cloud weaving, warp defect, other defects weft defect detection, identification pile surface fabric, yarn structure analysis and the like.
5
Future development trend of machine vision systems
1) the rapid development of embedded solutions, smart camera performance and cost advantages prominent, embedded PC will become stronger
2) Modular universal software platform and artificial intelligence software development technology platform will reduce personnel requirements and shortening development cycles
3) 3d vision will move toward more application scenarios
Letter to switch to AI students in school
[AI] complete AI self-learning course, the most detailed resource consolidation!
AI switch need to look at some of the articles
[PDF] to send the book Python programming from entry to practice
Python from entry to the master, the depth of learning and machine learning materials spree!
Press scan code tease returnees