{"id":{"repo_id":"mcmaster","oai_identifier":"oai:macsphere.mcmaster.ca:11375/32269"},"canonical_url":"https://search.ndltd.org/etd/mcmaster/oai:macsphere.mcmaster.ca:11375/32269","repository":{"repo_id":"mcmaster","name":"McMaster University","base_url":"https://macsphere.mcmaster.ca/server/oai/request"},"display":{"title":"Secure and Robust deep learning in computer vision","abstract":"In recent years, deep learning has become the foundation of modern computer vision applications, enabling machines to recognize objects, understand scenes, and make decisions based on visual data. However, deep neural networks can be vulnerable to security threats and unstable behavior, especially when exposed to adversarial inputs, poisoned training data, or complex real-world environments. This thesis presents three research efforts to improve the security and robustness of deep learning in computer vision. First, we propose a new defense method called Multi-Pronged Defense (MPD), which protects deep neural networks from backdoor attacks—hidden manipulations that cause models to behave incorrectly when triggered. MPD combines semi-supervised learning, balanced data sampling, and neuron suppression to effectively block various backdoor strategies across different datasets. Second, we design a novel attention mechanism for vision transformers that incorporates position-aware operations. This structure improves the model’s sensitivity to spatial patterns, similar to convolutional neural networks (CNNs), and achieves better performance than traditional self-attention while maintaining architectural flexibility. Third, we introduce the Balanced Object Detector (BOD), a new object detection framework that does not rely on feature pyramid networks (FPNs). By using consistent receptive fields and parameter sharing across detection branches, BOD achieves higher accuracy on small and medium objects and shows better generalization and resistance to adversarial attacks. Together, these contributions advance the development of deep learning models that are not only accurate, but also secure, stable, and reliable in real-world visual applications.","abstract_html":"In recent years, deep learning has become the foundation of modern computer vision applications, enabling machines to recognize objects, understand scenes, and make decisions based on visual data. However, deep neural networks can be vulnerable to security threats and unstable behavior, especially when exposed to adversarial inputs, poisoned training data, or complex real-world environments. This thesis presents three research efforts to improve the security and robustness of deep learning in computer vision. First, we propose a new defense method called Multi-Pronged Defense (MPD), which protects deep neural networks from backdoor attacks—hidden manipulations that cause models to behave incorrectly when triggered. MPD combines semi-supervised learning, balanced data sampling, and neuron suppression to effectively block various backdoor strategies across different datasets. Second, we design a novel attention mechanism for vision transformers that incorporates position-aware operations. This structure improves the model’s sensitivity to spatial patterns, similar to convolutional neural networks (CNNs), and achieves better performance than traditional self-attention while maintaining architectural flexibility. Third, we introduce the Balanced Object Detector (BOD), a new object detection framework that does not rely on feature pyramid networks (FPNs). By using consistent receptive fields and parameter sharing across detection branches, BOD achieves higher accuracy on small and medium objects and shows better generalization and resistance to adversarial attacks. Together, these contributions advance the development of deep learning models that are not only accurate, but also secure, stable, and reliable in real-world visual applications.","abstract_has_math":false,"creators":["Xu, Zhiwei"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":"Computing and Software","school":null,"contributors":[],"advisors":["He, Wenbo"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-11-19T05:57:29Z","subjects":["Computer Vision","Backdoor Defense","Object Detection","Vision Transformer"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://doi.org/10.71548/19475"],"render_values":[{"text":"https://doi.org/10.71548/19475","href":"https://doi.org/10.71548/19475","code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"http://hdl.handle.net/11375/32269","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"http://hdl.handle.net/11375/32269","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://macsphere.mcmaster.ca/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Amacsphere.mcmaster.ca%3A11375%2F32269","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["He, Wenbo"]},{"key":"dc:contributor.department","label":"Department","values":["Computing and Software"]},{"key":"dc:creator","label":"Author","values":["Xu, Zhiwei"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-08-29T18:15:14Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-08-29T18:15:14Z"]},{"key":"dc:date.issued","label":"Date","values":["2025"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Computer Vision","Backdoor Defense","Object Detection","Vision Transformer"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["http://hdl.handle.net/11375/32269","https://doi.org/10.71548/19475"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["In recent years, deep learning has become the foundation of modern computer vision applications, enabling machines to recognize objects, understand scenes, and make decisions based on visual data. However, deep neural networks can be vulnerable to security threats and unstable behavior, especially when exposed to adversarial inputs, poisoned training data, or complex real-world environments. This thesis presents three research efforts to improve the security and robustness of deep learning in computer vision. First, we propose a new defense method called Multi-Pronged Defense (MPD), which protects deep neural networks from backdoor attacks—hidden manipulations that cause models to behave incorrectly when triggered. MPD combines semi-supervised learning, balanced data sampling, and neuron suppression to effectively block various backdoor strategies across different datasets. Second, we design a novel attention mechanism for vision transformers that incorporates position-aware operations. This structure improves the model’s sensitivity to spatial patterns, similar to convolutional neural networks (CNNs), and achieves better performance than traditional self-attention while maintaining architectural flexibility. Third, we introduce the Balanced Object Detector (BOD), a new object detection framework that does not rely on feature pyramid networks (FPNs). By using consistent receptive fields and parameter sharing across detection branches, BOD achieves higher accuracy on small and medium objects and shows better generalization and resistance to adversarial attacks. Together, these contributions advance the development of deep learning models that are not only accurate, but also secure, stable, and reliable in real-world visual applications."]},{"key":"dc:description.abstractgeneral","label":"General Abstract","values":["This thesis explores how to make deep learning in computer vision more secure and robust. It introduces a defense method against backdoor attacks, a new attention mechanism that improves spatial sensitivity in vision transformers, and a reliable object detection architecture that performs well on small objects and under adversarial conditions. Together, these methods enhance the stability, generalization, and trustworthiness of visual learning systems."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Doctor of Philosophy (PhD)"]},{"key":"dc:title","label":"Title","values":["Secure and Robust deep learning in computer vision"]},{"key":"others:access_status","label":"Others Access Status","values":["open.access"]}]}],"canonical_facts":{"dc:contributor.advisor":["He, Wenbo"],"dc:contributor.department":["Computing and Software"],"dc:creator":["Xu, Zhiwei"],"dc:date.accessioned":["2025-08-29T18:15:14Z"],"dc:date.available":["2025-08-29T18:15:14Z"],"dc:date.issued":["2025"],"dc:description.abstract":["In recent years, deep learning has become the foundation of modern computer vision applications, enabling machines to recognize objects, understand scenes, and make decisions based on visual data. However, deep neural networks can be vulnerable to security threats and unstable behavior, especially when exposed to adversarial inputs, poisoned training data, or complex real-world environments. This thesis presents three research efforts to improve the security and robustness of deep learning in computer vision. First, we propose a new defense method called Multi-Pronged Defense (MPD), which protects deep neural networks from backdoor attacks—hidden manipulations that cause models to behave incorrectly when triggered. MPD combines semi-supervised learning, balanced data sampling, and neuron suppression to effectively block various backdoor strategies across different datasets. Second, we design a novel attention mechanism for vision transformers that incorporates position-aware operations. This structure improves the model’s sensitivity to spatial patterns, similar to convolutional neural networks (CNNs), and achieves better performance than traditional self-attention while maintaining architectural flexibility. Third, we introduce the Balanced Object Detector (BOD), a new object detection framework that does not rely on feature pyramid networks (FPNs). By using consistent receptive fields and parameter sharing across detection branches, BOD achieves higher accuracy on small and medium objects and shows better generalization and resistance to adversarial attacks. Together, these contributions advance the development of deep learning models that are not only accurate, but also secure, stable, and reliable in real-world visual applications."],"dc:description.abstractgeneral":["This thesis explores how to make deep learning in computer vision more secure and robust. It introduces a defense method against backdoor attacks, a new attention mechanism that improves spatial sensitivity in vision transformers, and a reliable object detection architecture that performs well on small objects and under adversarial conditions. Together, these methods enhance the stability, generalization, and trustworthiness of visual learning systems."],"dc:description.degree":["Doctor of Philosophy (PhD)"],"dc:identifier.uri":["http://hdl.handle.net/11375/32269","https://doi.org/10.71548/19475"],"dc:language.iso":["en"],"dc:subject":["Computer Vision","Backdoor Defense","Object Detection","Vision Transformer"],"dc:title":["Secure and Robust deep learning in computer vision"],"dc:type":["Thesis"],"others:access_status":["open.access"]},"updated_at":"2026-11-19T05:57:29Z"}