False-Recognition Target Extraction for Recognition Device Learning

The patent extracts only the false recognition targets from the set of all recognition targets to create the training data. By identifying targets that are likely to be misrecognized and using only those for training, the system achieves efficient learning without needing to train on all possible targets, thus reducing learning time while maintaining recognition coverage.

The patent applies different treatment to different recognition targets based on their local characteristics. Instead of uniform training on all targets, it identifies specific targets with high false recognition probability and applies targeted training only to those, optimizing the learning process by focusing resources where they are most needed.

The system extracts false recognition targets by analyzing recognition results and identifying which targets are most likely to be misrecognized. By using only these extracted targets for training, the system achieves effective reduction of false recognition rates without the computational overhead of training on all targets, thus maintaining reliability while improving learning efficiency.

Instead of performing complete learning on all recognition targets, the patent applies partial learning only to the subset of false recognition targets. This partial action is sufficient to address the false recognition problem effectively, achieving the desired reliability improvement with significantly higher learning efficiency.