36Kr|Innovative light field imaging scheme entered the field of industrial inspection. Deepvision got nearly 50 million RMB of pre-A financing
According to a source from 36Kr, Hangzhou Deepvision Technology Co., Ltd. (hereinafter referred to as "Deepvision"), which uses computer vision to develop industrial high-precision appearance inspection scheme, announced that it has obtained nearly 50 million RMB of pre-A round financing, which is led by Vertex Ventures and followed by Qualcomm venture capital. shuailin Wang, founder of the company, said: the financing will be mainly used for the R&D of full spectrum vision technology, and accelerate the rapid transformation of existing R&D achievements.
In recent years, Machine vision as a major branch of artificial intelligence technology has become a large-scale industrial trend. Among them, 2D vision, as an early developed field with relatively mature and stable technology, has a broad industry foundation in appearance detection. With the rise of high-precision manufacturing industry and the improvement of industry production standards, the application technology of machine vision needs new break through. However, due to the barriers in the industrial field and the limitations of working conditions, machine vision still has great development potential in the field of high-precision metal surface inspection, which is the highest standard of appearance inspection.
Specifically, metal parts, such as automobile parts, bearings, gears and other precision parts, are the basic parts of the whole industrial parts field with a huge scale. However, the low efficiency of manual quality inspection is a common problem in all industrial basic parts industries at present. This is seriously affecting the quality and productivity efficiency. As a tool to replace human eyes, machine vision has higher stability, accuracy and applicability, which is expected to make up the last link of the whole production process automation. At the same time, the contradiction between the increasing high requirements and importance of industrial detection and the backward detection methods provides opportunities for the application of machine vision technology. Shuailin Wang believes that replacing low efficiency with high efficiency and making good use of technical advantages such as distributed computing, artificial intelligence and big data will bring changes to the field of industrial inspection and even the whole industrial production.
Founder shuailin Wang said Deepvision was established in June 2017 and has been focusing on technology research and development and product design. The company has developed a number of solutions for high-precision appearance inspection. At present, it has been implemented in traditional auto parts, aviation parts, new energy, textile, 3C and other scenes, serving more than 50 customers, and has signed contracts with many industry first-line customers.
Shuailin Wang, founder of Deepvision, said that the technical route of Deepvision is obviously different from most companies on the market. Deepvision has the R&D and design capabilities of the whole technology chain, including camera design and development, edge computing, image algorithm, light path design of light field, artificial intelligence algorithm, automation equipment design and many other core technologies, It can achieve higher detection efficiency and one-time pass rate, and greatly reduce the false detection rate and missed detection rate, Specifically:
In the visual design, the most crucial part is probably the optical inspection. The focus of the industry is how to observe a wide range of defects under LIMITITED lighting conditions. At present, most of the machine vision companies are using the imaging effect under some kind of illumination, which is an isolated problem under some static conditions in abstract, so as to guide the design. But the problem is that the visual inspection equipment faces a huge space to describe the defect imaging effect. The incremental thinking is not only time-consuming and labor-consuming, but also difficult to cover all scenes. Therefore, depth vision integrates all the known defect types, carries out complete analysis and test demonstration, and obtains systematic common analysis results, so as to obtain the imaging effect of defects under different light field structures, which is a kind of "subtraction" thinking. In the actual landing scene, there are many uncertain factors on the surface of precision metal workpiece, which will affect the detection results, such as the interference of industrial oil on the surface, the speed requirement of ultra-high speed online, the jitter of the conveyor, etc. In the case that many other laboratory technologies are still in the theoretical stage, Deepvision has achieved more than 99% defect type detection coverage and about 98% overall measured qualification rate on specific equipment through continuous testing and verification on the first line, and the detection accuracy is also in the micron level.
On the premise of establishing the underlying image and database, computing power is the key to determine the efficiency of equipment detection, and it is also the innovation of depth vision. Founder shuailin Wang mentioned unlike the traditional GPU accelerated centralized computing route. Deepvision adopts distributed computing based on FPGA. The obvious advantage of this is that the computing power is shared. The system can detect the target multiple times in different lighting scenes, thus greatly improving the detection efficiency. The practice shows that the one-time over inspection rate of the production line is 13-18% higher than that of the peers, and the false inspection rate is reduced by more than 10%.
Deepvision is expanding and will always look for opportunities to expand and grow both financially and technically. At this time, the company has completed the first set of delivery for many industry head users and will start next round large-scale promotion. Based on the feedback from the customers, the return on investment cycle is basically controlled within 1-3 years, and one equipment can replace about 5-10 labor force on average, which is related to the use of production line and equipment.
Managing partner Zhi Xia from Vertex Ventures said: “The detection application accounts for more than half of the total vision field. In the detection application field, metal and glass detection are the two of most difficult. Metal materials are highly reflective, defect collection is difficult, and the scene is scattered and diverse, so the requirements for detection are especially high. Deepvision combines artificial intelligence and machine vision technology to effectively detect high reflective and high curvature metals and solve customers' pain points. Deepvision has been recognized by the industry giants. I believe that the application of visual inspection in the industrial field will be more and more abundant, and the development space of Deepvision is huge. "
"At present, artificial intelligence technology has been widely used in many vertical fields," said Song Mao, chief investment officer of Qualcomm venture capital. Deepvision has made use of its technical breakthrough in visual imaging and deep learning algorithm optimization in metal surface detection to provide engineering solutions for the industry, which has been widely recognized by industry users. "