Industrial evolution of machine vision: semiconductor opens the imagination space

Industrial evolution of machine vision: semiconductor opens the imagination space

Our correspondent Luo Yi Qi reports from Shenzhen.

As artificial intelligence technology penetrates into daily life on a large scale, more and more machine vision companies are attracting attention in the industrial field, especially in the mobile phone manufacturing sector which occupies a huge market.

Recently, in an interview with reporters, a number of machine vision company founders said that security applications have emerged many unicorns, industrial scenes, especially in the quality testing link, is becoming the next direction to be targeted. Especially in the background of Guangdong, Hong Kong and Macau Dawan area, the developed manufacturing industry in the Pearl River Delta is speeding up the integration with the industrial, academic and research capabilities concentrated here.

Especially in recent years, with the strong support of the semiconductor industry, some machine vision companies are "aiming" at semiconductor packaging and other links, and are embarking on the next technological breakthrough.

Industrial testing soil

From the perspective of computer vision, the pace of industrialization landing begins with human relations.

On October 25, Lu Yanxia, Senior Research Manager at IDC in China, told reporters that in the relatively mature application market, the software supply module, static face recognition, public area entrance guard and other areas of the security sector are growing rapidly; but segmentation into dynamic face recognition, such as snapshot, still needs to be certain. Time iteration. In addition, in the field of intelligent transportation, the demand caused by urban congestion has led to a larger scale in the past 1-2 years.

Starting in 2018, the specific application scenarios of mobile terminals and the Internet began to become a significant growth point, the latter including intelligent building, 3D fitting and other scenes. "Next, driven by Ali, Jingdong and other innovative companies, the new retail AI application scenario is expected to become the next hot spot in 2018-2019. In addition, medical image recognition, autopilot and other market space is also very large.

In the field of machine vision, Lu Yanxia believes that from a macroscopic point of view, its overall algorithm and application has not yet reached the stage of large-scale explosion. First of all, it is related to the degree of automation.

"The establishment of computer vision model requires a large number of data resources, from the current domestic manufacturing industry as a whole, the degree of networking is not enough. But in the industrial application scenario, the accuracy of AI algorithm is very high, otherwise it will bring security risks. As a whole, she said, the outbreak of machine vision used to create scenes in China is strongly related to the degree of industrial Internet advancement, or will take 2-5 years to develop gradually.

Lu Yanxia analyzed that the manufacturing industry is very different, has its own industry rules, but also needs a deep understanding of the industry. "As far as I know, the average accuracy in real-world scenarios is about 80-90% for industrial manufacturing, and 98-99% for individual ideal states. But more should be in data acquisition and training model phase. Therefore, it will be used in the manufacturing of products that are not so high in quality and accuracy. "

Of course, the advantage of the Pearl River Delta is that the manufacturing scene is rich, and the development of intelligent manufacturing related new factories are gradually built with policy support. This is also the background of the emergence of a number of focused machine vision technology landing companies in the Dawan area, of which the mobile phone industry chain is most concerned because of the huge market space.

On October 19, Jiang Yong, chairman of High-tech, told 21st Century Economic Reporter that industrial automation is divided into two major areas, the first is industrial production equipment, the second is industrial testing equipment. It is the latter that focuses on technology. "The second area means that production automation is basically complete, and such companies have a large output, a strong strength, and a good foundation. The mobile phone related field is a large number of representative industries with strong iteration, and the profit margin is very good.

The demand for machine vision in detection equipment is urgent, and it is also related to the high cost of human detection. From the trend, the battery, cover board and screen in the mobile phone industry chain have been given priority attention.

"For example, in 2017, 400,000 workers will be required to test the screen, but the screen detection is more eye-damaging and irreparable. So in this scenario, machine substitution is a trend. Jiang explains that China accounts for 50% of the world's mobile screen capacity and nearly 70% of its batteries.

Visual technology focuses on mobile phone cover defects detection. Liu Ying, founder and CEO of the company, told 21st Century Economic Reporter that the industry tested in this field is about 19.6 billion yuan.

Semiconductor packaging opportunities

Invariably, the above executives expressed their enthusiasm for the participation of semiconductor industry. This is related to policy support, and from a technical point of view, it means higher level competition.

"Our research and development will continue to move in the direction of challenging, high-tech barriers, the actual landing of course to take into account the problems of making money and easy landing, such as buttons, battery testing and so on. But we are more optimistic about the semiconductor industry. Liu Ying told reporters.

Jiang Yong also pointed out that the detection of semiconductor packaging stage is the direction of the company's attention, and has begun to layout, although in the domestic industrial manufacturing process, semiconductor packaging is not yet the mainstream market, has not yet reached the "explosion point", but "with the support of government policy, perhaps 1-2 years later will start to bring opportunities. We're going to lay it out in advance, because it's probably a year before the technology is ready, and we want to catch up in a year.

Liu Ying further explained, "Semiconductor is a very long cycle, demanding industry, of course, we dare not enter easily, the detection of semiconductor is not a company can easily do, in the future we will consider seeking partners to intervene together."

With the support of the policy, the domestic semiconductor industry ushered in a new round of heat this year, and the enthusiasm of industry acquisition and acquisition is also rising. However, compared with the very mature mobile phone industry chain in China, the semiconductor packaging field is actually developing slowly, such machine vision companies still have a long way to go.

Liu Ying is well aware of the need to reach the nanometer level, both in terms of precision and configuration of optical components. This will require more than software, as well as machinery and electronic control.

"To match higher precision semiconductor packaging, optics and vision algorithms are the core." Jiang Yong also said that for this reason, the machine vision company's iteration will also be based on the "generation of equipment" as a unit, rather than just "a batch of equipment.

However, at present, the scarcity of AI talents is a universal problem. Although the linkage of industry, University and research in Dawan District can bring better basic resources, it is necessary to be alert to the problem of industry awareness.

For example, Liu Ying pointed out to reporters, "Actually, there are very few AI talents, but the industry is now relatively good, and there are some ready-made open resources, sometimes give young people the illusion that taking a few photos to train the results, is to understand the AI algorithm, but AI research its own structure, resources, data. In fact, it is very particular. So there is still a shortage of truly proficient people. In addition, GPU and other hardware are still using foreign products.

But she's confident, "As long as there's an atmosphere in this industry, it's not particularly worrying. It's just a matter of thinking about where to focus, not just seeing the industry hot.

(Editor: Zhang Weixian)

Waonews is a news media from China, with hundreds of translations, rolling updates China News, hoping to get the likes of foreign netizens