In July 2019, Dongsheng (Suzhou) Intelligent Technology Co., Ltd was established. At the beginning of 2020, tens of millions of angel rounds of financing will be completed. In the words of Han Xu, CEO of Dongsheng, Dongsheng focuses on artificial intelligence and machine vision as always, and intelligent manufacturing "Zhi" is the core. It uses artificial intelligence technology to serve the industry after fitting the huge, discrete and multi-dimensional data of the industry. What Dongsheng does is to install a pair of technological "smart eyes" for the factory. A pair of sound monitoring technology master "wise ear" and a production line big data analysis technology of the super "brain", so that the industry becomes more intelligent.
Out of curiosity about such a rapidly developing machine vision enterprise, we interviewed Han Xu, CEO of Dongsheng (Suzhou) Intelligent Technology Co., Ltd. Han Zong on their own enterprise and industry needs, the development of thinking, hope to provide you with a unique perspective.
Han Xu, CEO of Dongsheng (Suzhou) Intelligent Technology Co., Ltd
"The COVID-19 is like a magnifying glass, exposing potential problems in all industries. The global market has been hit hard, and the Chinese industry has been hit hard, slowing down the progress of the whole industry," Han said. But as the so-called "crisis in organic", once the epidemic is over, industry will become the backbone of economic recovery, and intelligent manufacturing will become the core of the backbone of power. For the machine vision industry is also will be a greater opportunity, its development will be "invincible". The epidemic has caused many manufacturing enterprises to press the "slow down" or even "pause" button, exposing the shortage of workers in factories."
Through this, domestic enterprises will speed up the transformation of intelligent manufacturing, as an essential intelligent process of AI machine vision east sound technology will also develop rapidly. As an enterprise with machine vision and artificial intelligence algorithm as its core and a team with research and development as its core, the epidemic has not only made us stagnate, but also made a leap forward in our research and development. We have taken advantage of the epidemic to focus on research and development.
Based on the technical talents and think tank team of Dongsheng, we have broken through the core difficult application of artificial intelligence:
(1) The small sample data training is used for the lack of pictures of individual defect types and the incomplete understanding of defect types of products. Generative adversarial network and other algorithms are used to simulate rich data and make up for the shortage of defective samples.
(2) implementation across product model is available in the product upgrades or product model change, parallel for different products or similar material, etc., through the transfer of learning to big data model can be trained by the migrated to multiple areas , only a small data field in the artificial intelligence in the industrial application of generalization ability and adaptability.
(3) To achieve a single positive sample data training product initial production without defect pictures, the use of our algorithm can greatly reduce the cost and time, only training OK pictures, through a lot of learning and analysis, can detect the defects on the defective products, solve the pain points of no defect pictures.
The realization of high adaptability to industrial environment is not affected by the industrial environment, not affected by the decline of light source, camera, lens and other hardware performance, and still maintains a high degree of accuracy within a certain range of changes in the external environment.
Handdleai, a core algorithm platform for artificial intelligence, will be at the show
Mr. Han explained that we have been in the machine vision industry for many years. Although this year is our first appearance in VisionChina (Shanghai), we are fully prepared and ready for it. VisionChina (Shanghai) will be on display during 2020: Handdleai, the core algorithm platform of Dongsound artificial intelligence, detection equipment and actual cases of various scenes such as artificial intelligence mobile phone cover glass detection equipment applied in 3C industry and 3C industry display equipment combining 2D detection and 3D. Please stay and look forward to it.
We have technical and manufacturing industry many enterprises to establish the relationship between the depth of cooperation, such as: new energy soft packing lithium ion battery appearance defect detection, detection rate of production line at present the traditional algorithm of the project: miss rate of 0.03%, after application east sound AI software testing to 0.01% (due to the amount of data updating in improve detection rate). For the appearance inspection of semi-finished products of new energy batteries, the production line is currently visual inspection by personnel, with low accuracy and speed. Each line has 10-15 people per shift and the 12-hour system, which still cannot meet the production demand. After the introduction of AI detection scheme in the production line, the detection accuracy is increased by 5% and the speed is increased by 6-8 times compared with the previous detection.
Semiconductor field LED lamp bead detection, product size between 1mm-5mm. CT has high requirements (36K/ h), random defect location, complex defect feature extraction and diverse product types. There are 26 common defect types, among which foreign body, surface scratch, surface glue, lack/less glue, bubble and so on are the key detection defects. The current situation of the industry manual microscope sampling, most of the important defects can not be automated detection, east sound accuracy reached 99.98%.
For any industrial product, defects are related to the quality of the product, and the quality is related to the survival of the enterprise. An industrial product has dozens of materials at least, and thousands of parts at most. Each process needs to carry out appearance defect detection more or less. With the increasing precision of products, defect detection has become more and more difficult. Now the current manual visual inspection and traditional vision can no longer meet the requirements, and the number of unqualified products flowing to the market has increased sharply, which urgently needs to be changed. The current technology of Dongsheng Technology can solve the above problems. In the aspect of machine vision, Dongsheng has mature Handdleai (2D+3D) common supplies, which can be directly used for defect detection of images. At the same time, it can provide soft + hard solutions (including optical system and customized demand development algorithm). Through our "eyes" accumulation of industry big data and our "brain" international algorithm ability plus the organic integration of computing power, so as to promote the transformation and upgrading of the whole factory.
Dongsheng and 5G, artificial intelligence technology
The arrival of 5G has realized the interconnection of everything and created favorable conditions for the industrial interconnection belt. Vision is the most important eye of the industry, and 5G will certainly make the coordination between brain and eyes more smoothly. Throughout the industry at present most of the visual inspection are marginal computation, the coming of the 5 g will give customers more choice, you can continue to edge calculation can also be private or public clouds, from the point of view of communication is currently not implemented on the cloud of work, including industrial image can reach dozens or even hundreds of megabytes size, short transmission speed requirements of production line.
Summing up how Dongsheng's technology will integrate with 5G, Han said: "5 g arrival let cloud implementations may, greater bandwidth, faster transmission speed, lower communication latency, higher reliability and other advantages, more and more customers to our consulting relevant business, at the same time, we are actively to deploy cloud detection technology, we will launch more artificial intelligence cloud detection scheme, create greater benefits for industrial, please stay tuned."
In the future, companies may rely more than ever on embedded vision systems
Finally, we have a warm discussion with Mr. Han about the increasingly prominent technology fusion today. What technology fusion does he think is the most important for machine vision? At the same time, what does he think are the potential trends or new directions for the machine vision industry?
Mr. Han shared that in addition to the proven deep learning technology, in the future, Dongsheng will explore in the fields of 3D vision +AI, 3D vision +2D vision +AI, image information + sound information +AI, and embedded vision +AI technology. We firmly believe that embedded visual applications will play a great role in Industry 4.0 intelligent manufacturing, with advantages of rapid deployment, low cost, rapid response, rapid feedback, etc., and will be widely used in the manufacturing process and testing link of the factory production line.
What would a smart factory look like? First, it will provide extensive interoperability and a high level of communication capability for machines, equipment, sensors and people. Secondly, it will have high information transparency. The system creates a virtual map of the physical world through sensor data, making the information situational. Third, the decision-making behavior of the smart factory will be highly decentralized, and the physical network system will operate autonomously as much as possible. Fourth, these factories will provide a high level of technical assistance, systems that can help each other solve problems, make decisions and help humans complete difficult or dangerous tasks. In almost all cases, enterprises will rely more than ever on embedded vision systems.
The AI intelligent camera of Dongshen is a small AOI embedded vision device that integrates industrial camera + artificial intelligence algorithm + image processing algorithm + hardware speed optimization scheme + optical scheme. In addition to realizing OCR, positioning and relatively simple detection functions, the layout can collect data on the production line, obtain the manufacturing process of the entire production line, and use deep learning algorithm to extract process features to establish the mapping relationship between image information and process, improve process and increase productivity.