Zebra Aurora ™ DEEP LEARNING Deep Learning Development Environmental Tool
Validge

Automatic AI exterior inspection in the company is low at low cost Easy to develop just by reading the image This add -on product is Aurora ™ Vision to respond to problems that are difficult to solve with conventional image processing algorithms. It is an industrial quality deep learning tool that can be used to further strengthen Studio. The actual realization by combining features and abnormal detection, advanced optical character recognition (OCR) technology, and deep learning function The system can recognize the use of image information in the field, and to evolve the system accordingly. Aurora ™ DEEP LEARNING is a large -scale new newly designed and optimization designed and optimized by ZEBRA's research team for industrial inspection systems. I am using a ralnetwork. ･ Learn from a small number of samples In general applications Learn with 20 to 50 images for learning I can. Those who have more sample images Is better, but aurora ™ Deep Learning software Lord from a limited training set Learn the key characteristics and have effective trays A new system is new for ning Equipped with a function to generate thousands of cappules I'm sure. ･ Operation with GPU and CPU For effective training You need the latest GPU. In the actual machine, either GPU or CPU Can be used. With a GPU, compared to CPU Usually 3 to 10 times faster. (Object classification is an exception, CPU But it's the same speed). ･ Guarantee high performance Typical learning time on GPU is usually 5-15 minutes. Progenon time is tools and hardware Depending on the image, 5ms per image It is about 100ms. Our industrial inference engine Use Aurora ™ Weaver So, guarantee the best performance increase. Application Examples ･ Features (mode with teachers) Specify pixels corresponding to defects in the learning image (labeled) It is a method to learn. It can be detected accurately. ･ Abnormal detection (mode without semi -teacher) Learning is simpler. Defects are not strictly defined, but good Learn only with a sample and look for a deviation from a good product. Aurora ™ Deep Learning offers two methods. The first one The second is one class of all parts of the input image, using image reconstructive technology. Execute the classification. If you need a high -precision defective heat map, calculate We recommend the first method at the expense of time. ･ An object classification Traditional image processing can also be difficult classifications. ･ Instance segmentation The instance split is a single or multiple objects in the image It is used to identify the position of the) and divide and classify. Special Unlike the encoding, to detect individual objects, Even if the vehicle comes in contact or overlaps, it can be separated to come. ･ Point glocation Specific shapes, features that can be identified as points on the input image, Or look for a mark. With conventional template matching It may be compared, but this tool is multiple samples Because it is learned in, Robasty for major fluctuations in the object Is high. ･ OCR (Optical Character Recognition) Character recognition (OCR) machine included in Aurora ™ DEEP Learning Noh is complicated and uneven background, blurred characters, damaged characters, Due to distorted characters, unclear characters, reflected metal surface, etc. Difficult character recognition so that conventional OCR technology cannot be used The answer to the logect. This tool uses thousands of different image samples in advance. It comes with a neural netet that can be used immediately vinegar. This tool is immediately, even if you handle very difficult cases Up to 97% accuracy can be achieved, the user is a machine Rough in just a few steps without the need for vision expertise You can create an OCR application.