NTT COMWARE English Site


NTT Comware's Image Recognition AI "Deeptector®": New Functions Available including Positive Case Reasoning
-- Easy detection of defective products by AI learning small data of non-defective products --

NTT Comware Corporation (Head office: Minato-ku, Tokyo; President: Satoshi Kurishima, hereinafter NTT Comware) releases the new functions of "Deeptector®*1", the "positive case reasoning" and "portability of learned models" on October 1, 2018.

* "Deeptector®", the image recognition AI, is a service commonly used for various NTT Comware's technologies*2 comprising of NTT Group's "corevo®".

1. Positive Case Learning
The Deeptector's detection patterns of image recognition include "detection", "classification" or "level". The customers select a recognition pattern suitable for their requirements, even combining multiple patterns. On the other hand, Japanese manufacturing industry hesitated to use AI since conventional AIs, which would be required to learn a large number of negative examples (abnormal conditions or defective products), take time to sufficiently collect images to learn. This is because the quality of their products is quite high and little defects occur on the production line.

One of the newly added pattern recognition is "positive case learning", which makes AI possible to recognize negative examples (abnormal state, defective items) by learning only a small data of some positive examples (normal state, non-defective products). Therefore, the image recognition AI can be used in a quicker and easier way. Especially, it is effective for the cases that customers have little data of defective products, the ratio of products with defects is low or they have a small lot production with various type of products.

[Figure 1. "Positive Case Learning" (example of defects of a plastic bottle cap)]

Figure 1.

2. Portability of Learning Model
Although the learning model was not compatible between the cloud and software version of "Deeptector" the image recognition AI, the same learning model becomes possible for both cloud and software versions thanks to the new function "portability of learning model". Since learning models generated in a cloud environment running on high-end GPUs can also be used in the software version on a customer's server, it is a time-saving way for re-learning after AI is applied. It is effective for cases if the same learning model is used on GPUs in factories in different locations throughout Japan.

[Figure 2. Portability of Learning Model]

Figure 2. Portability of Learning Model

* NTT COMWARE will exhibit the new version of AI "Deeptector" in the following events.
* "The 5th Next Generation Agricultural EXPO" October 10 (Wed) - 12 (Fri), 2018 at Makuhari Messe
* "The 2th Japan IT Week Fall, AI/ Business Automation Exhibition" October 24 (Wed) - 26 (Fri) 2018 at Makuhari Messe


The information shown in this page is the latest as of the day of the release. It is subject to change without further notice. Thank you for your kind understanding and consideration. PageTop