AI anomaly detection that learns only from good product images | No defective samples needed
DeepAge Sensor
Learn the standard of "normal" from good product images and detect those that deviate. It can also identify defects that have not been seen before. CPU operation.
In processes where defects rarely occur, it is impossible to collect defective product samples. DeepAge Sensor's unsupervised anomaly detection learns the "normal" standard using only good product images and detects images that deviate from that standard. It does not ask for "hundreds of defective product images." Inspection operations can start from the early stages of implementation, and even defects that have never been seen before can be identified as "deviating from normal." Inference operates on CPU (no GPU required), and integration into the equipment is done with just one DLL. Later, if defective product images accumulate on-site, it can transition to a supervised model with the same API structure. For processes with insufficient samples, accuracy can first be verified through free model creation.
Inquire About This Product
basic information
【Learning Method】Unsupervised anomaly detection (learning normal standards with only good images) 【Detection Principle】Detecting images that deviate from normal with an anomaly score (capable of handling unseen defects) 【Inference Speed】CPU 12ms/image (no GPU required, ONNX Runtime included) 【Model Format】.uai (includes decision threshold within the model, no threshold management needed on the device side) 【Visualization】Displays "where it deviated from normal" with a contribution heatmap 【Transition】When defective images are collected, switch to a supervised model (API structure remains the same)
Price information
Direct sales: 250,000 yen/camera/year〜/First model creation free
Price range
P3
Delivery Time
P3
※The initial model creation is free (accuracy report in as little as one week).
Applications/Examples of results
- Launch of visual inspection in processes with rare defects (no defective product samples needed) - Initial phase is unsupervised, transitioning to supervised in a gradual operation - Detection of unknown types of defects as "deviations"
Detailed information
-
8 years of operational performance and key specifications in the manufacturing site.
-
Unsupervised anomaly detection learning "normal" only from good product images.
-
Learning is in the cloud, inference is on-site with CPU (no GPU needed).
-
Send only good product images → Learning → Receive DLL → Integration (first time free)
-
Applicable to six industry categories (as deviation detection)
catalog（2）
Download All Catalogs
Company information
We provide services that utilize data. The value generated from data has been increasing day by day. While there may be a focus on chasing numbers in sales targets and marketing, the importance of processing large amounts of data with computers to create new value has not been emphasized as much. Technology is evolving every day. We have become capable of gathering various data in the cloud and processing large volumes of data. However, even with an understanding of such technologies and their value, practical implementation is still not sufficiently realized. We will accurately understand the current issues, sincerely consider whether artificial intelligence and new technologies can solve them, and communicate our findings to the world in the most optimal way.
![[VE Evaluation] Weather Risk Management Mobile 'KIYOMASA PRO'](https://image.www.ipros.com/public/product/image/2110087/IPROS2916142205131244335.png?w=280&h=280)
