Automatic detection of prohibited items is a critical but difficult task during aviation security. Manual detection of such items is a timeconsuming process that is also limited by the examination capacity of the security inspector. To overcome these constraints, several researchers have proposed deep learning solutions to identify contraband data contained within baggage X-ray imagery. However, when trained on the imbalanced data that is frequently encountered in real-world aviation screening, the performance of these models suffers significantly. Towards this end, this paper proposes the coupling of various imbalanced learning strategies that can be used to augment traditional threat detection models and enable them to effectively learn the extremely imbalanced distribution of normal and threat object categories. The proposed approach is validated on three public datasets, namely SIXray, OPIXray, and COMPASS-XP, where it achieved the performance improvement of 9.52%, 11.32%, and 10.98%, respectively, on all three datasets in terms of mean intersectionover-union as compared to the state-of-the-art threat detection frameworks.