Simultaneous Classification of Objects with Unknown Rejection (SCOUR) Using Infra-Red Sensor Imagery.

Area of Science:

• Computer Vision
• Machine Learning
• Defense Technology

Background:

• Infra-red target recognition is crucial for defense and security.
• Existing classifiers struggle to reject unknown objects without misclassification.
• Need for robust systems that identify known targets and reject novel threats.

Purpose of the Study:

• Enhance pre-trained classifiers to detect and reject unknown classes.
• Maintain classifier performance on known classes.
• Develop a method that does not require out-of-distribution (OOD) data for training.

Main Methods:

• Introduced a secondary regression network to work with a primary classifier.
• Combined primary classifier confidence with secondary network's class-conditional score.
• Utilized a Bayesian framework for improved separation of known and unknown objects.

Main Results:

• Demonstrated effectiveness on CIFAR-10 and a medium-wave infra-red (MWIR) dataset.
• Outperformed state-of-the-art methods in rejecting unknown target types.
• Maintained accurate classification of known targets.

Conclusions:

• The proposed method effectively enhances infra-red target recognition systems.
• Successfully separates unknown objects from known classes without OOD training data.
• Offers a promising solution for defense and security applications requiring robust target identification.