Date of Award
5-1-2024
Degree Name
Master of Science
Department
Electrical and Computer Engineering
First Advisor
Tragoudas, Spyros
Abstract
This research proposes a novel method to identify the YOLO object detection network failures. The proposed method employs a secondary neural network to predict misdetections (localization and classification error) based on the features extracted from the YOLO network. Moreover, to make the secondary network lightweight by selecting important features for a target class, a recursive feature elimination-based method is proposed. As a result, the computational cost is reduced without compromising the accuracy. Four of the most frequently occurring classes in the COCO dataset were taken into consideration when training the secondary network for the experimental evaluation. When a single class was taken into consideration, the proposed failure detection approach attained an accuracy of 89.79%; this is a 16% improvement in accuracy over the existing method. A 52% reduction in inference time and an accuracy of 88.90% were obtained with the feature selection approach. Additionally, the proposed failure detection framework was assessed by taking into account several classes at once, and excellent accuracy was noted.
Access
This thesis is only available for download to the SIUC community. Current SIUC affiliates may also access this paper off campus by searching Dissertations & Theses @ Southern Illinois University Carbondale from ProQuest. Others should contact the interlibrary loan department of your local library or contact ProQuest's Dissertation Express service.