The “Deep Learning Technologies for Defect Classification” of ITRI CITC, adopts the patented Deeply-Fused Branchy Network architecture to take both accuracy and inference latency into account to fulfil industrial strict requirements. This technology has been transferred to many domestic companies, and the implementation results are satisfied.
Deep Learning research has achieved many substantial breakthroughs in recent years, making the scope of computer vision application, especially the field of image classification, no longer limited by the bottleneck of handcrafted features, but can be quickly extended to deal with many kinds of classification problems.
The “Deep Learning Technologies for Defect Classification” of ITRI CITC adopts the patented Deeply-Fused Branchy Network architecture to take both accuracy and inference latency into account to fulfil industrial strict requirements. In addition, given an already trained model, miss rates of the class of interest can be further suppressed without the need of re-training model. Such flexibility can greatly facilitate industrial production deployments.
Furthermore, to achieve the same accuracy, the high-efficiency data screening and labeling techniques developed by ITRI CITC could analyze model’s predictions to figure out which examples are more effective in training model, thereby greatly reduces the number of samples that need to be manually labeled. Applications in many different fields such as security, surveillance, machine vision, and AOI, etc. can benefit from the technologies a lot.
The technologies have been transferred to domestic companies, including the manufacturing industry of semiconductor, wafer inspection, PCB, PCBA, and so on. Empirical evidences confirmed the effectiveness of the aforementioned solution.