{"id":{"repo_id":"stellenbosch","oai_identifier":"oai:scholar.sun.ac.za:10019.1/136067"},"canonical_url":"https://search.dev.ndltd.org/etd/stellenbosch/oai:scholar.sun.ac.za:10019.1/136067","repository":{"repo_id":"stellenbosch","name":"Stellenbosch University","base_url":"https://scholar.sun.ac.za/server/oai/request"},"display":{"title":"Development of a ML Algorithm to Improve Radar Detection of Golf Balls Using MFCW Radar","abstract":"This thesis addresses the challenge of reliably detecting small targets, specifically golf balls, using Multi-Frequency Continuous-Wave (MFCW) radar in environments contaminated by significant noise and clutter. Although radar systems are effective at tracking targets, the performance of a radar system is heavily constrained by unwanted returns from external sources, such as fans, which often generate Doppler frequencies comparable to the target return signal. A comprehensive MFCW radar system simulation was developed, incorporating realistic physics and detailed models of critical interference targets. This enabled the design and testing of various detection algorithms. This thesis compares Constant False Alarm Rate (CFAR) detectors (including CA-, OS-, VI-, TM-, CHA- and WAI-CFAR) against modern Machine Learning (ML) techniques, including multinomial and binary Logistic Regression (LR) and Gradient Boosted Tree classifiers (XGBoost). For the ML models, a bin specific classification approach was explored, which utilised features which were engineered from statistical properties and signal peak characteristics. The models were trained on data collected from four international golf driving ranges. Results confirmed that classical CFAR detectors struggled in this cluttered environment, as seen by the false positive rates (FPR) which far exceeded the threshold that each detector was designed for. In contrast, the ML techniques, specifically XGBoost, which was used in a bin specific classifier, performed significantly better. The best model obtained a Recall (Probability of Detection) of 88.84%, with a False Positive Rate (FPR) of 0.431%. While the best performing CFAR detector, OS-CFAR obtained a recall of 81.42% with an FPR of 0.635%. These results indicate that ML models are capable of learning complex features beyond what is capable with CFAR detectors, allowing for more reliable detection of targets in environments with a low Signal to Clutter Ratio (SCR).","abstract_html":"This thesis addresses the challenge of reliably detecting small targets, specifically golf balls, using Multi-Frequency Continuous-Wave (MFCW) radar in environments contaminated by significant noise and clutter. Although radar systems are effective at tracking targets, the performance of a radar system is heavily constrained by unwanted returns from external sources, such as fans, which often generate Doppler frequencies comparable to the target return signal. A comprehensive MFCW radar system simulation was developed, incorporating realistic physics and detailed models of critical interference targets. This enabled the design and testing of various detection algorithms. This thesis compares Constant False Alarm Rate (CFAR) detectors (including CA-, OS-, VI-, TM-, CHA- and WAI-CFAR) against modern Machine Learning (ML) techniques, including multinomial and binary Logistic Regression (LR) and Gradient Boosted Tree classifiers (XGBoost). For the ML models, a bin specific classification approach was explored, which utilised features which were engineered from statistical properties and signal peak characteristics. The models were trained on data collected from four international golf driving ranges. Results confirmed that classical CFAR detectors struggled in this cluttered environment, as seen by the false positive rates (FPR) which far exceeded the threshold that each detector was designed for. In contrast, the ML techniques, specifically XGBoost, which was used in a bin specific classifier, performed significantly better. The best model obtained a Recall (Probability of Detection) of 88.84%, with a False Positive Rate (FPR) of 0.431%. While the best performing CFAR detector, OS-CFAR obtained a recall of 81.42% with an FPR of 0.635%. These results indicate that ML models are capable of learning complex features beyond what is capable with CFAR detectors, allowing for more reliable detection of targets in environments with a low Signal to Clutter Ratio (SCR).","abstract_has_math":false,"creators":["Holtshausen, Damian Luke"],"institution":"Stellenbosch : Stellenbosch University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Grootboom, L. L.","Steyn, W.","Niesler, T."],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-03","date_published":"2026-03","updated_at":"2026-07-24T04:40:09Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholar.sun.ac.za/handle/10019.1/136067","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Grootboom, L. L.","Steyn, W.","Niesler, T."]},{"key":"dc:contributor.other","label":"Dc Contributor Other","values":["Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering."]},{"key":"dc:creator","label":"Author","values":["Holtshausen, Damian Luke"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-04-21T13:02:42Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-04-21T13:02:42Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-03"]},{"key":"dc:publisher","label":"Institution","values":["Stellenbosch : Stellenbosch University"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"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":["https://scholar.sun.ac.za/handle/10019.1/136067"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (MEng)--Stellenbosch University, 2026.","Holtshausen, D. L. 2026. Development of a ML Algorithm to Improve Radar Detection of Golf Balls Using MFCW Radar. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/4937e1fc-e3d8-4bf4-9cfe-98ea33d8af2c"]},{"key":"dc:description.abstract","label":"Abstract","values":["This thesis addresses the challenge of reliably detecting small targets, specifically golf balls, using Multi-Frequency Continuous-Wave (MFCW) radar in environments contaminated by significant noise and clutter. Although radar systems are effective at tracking targets, the performance of a radar system is heavily constrained by unwanted returns from external sources, such as fans, which often generate Doppler frequencies comparable to the target return signal. A comprehensive MFCW radar system simulation was developed, incorporating realistic physics and detailed models of critical interference targets. This enabled the design and testing of various detection algorithms. This thesis compares Constant False Alarm Rate (CFAR) detectors (including CA-, OS-, VI-, TM-, CHA- and WAI-CFAR) against modern Machine Learning (ML) techniques, including multinomial and binary Logistic Regression (LR) and Gradient Boosted Tree classifiers (XGBoost). For the ML models, a bin specific classification approach was explored, which utilised features which were engineered from statistical properties and signal peak characteristics. The models were trained on data collected from four international golf driving ranges. Results confirmed that classical CFAR detectors struggled in this cluttered environment, as seen by the false positive rates (FPR) which far exceeded the threshold that each detector was designed for. In contrast, the ML techniques, specifically XGBoost, which was used in a bin specific classifier, performed significantly better. The best model obtained a Recall (Probability of Detection) of 88.84%, with a False Positive Rate (FPR) of 0.431%. While the best performing CFAR detector, OS-CFAR obtained a recall of 81.42% with an FPR of 0.635%. These results indicate that ML models are capable of learning complex features beyond what is capable with CFAR detectors, allowing for more reliable detection of targets in environments with a low Signal to Clutter Ratio (SCR)."]},{"key":"dc:title","label":"Title","values":["Development of a ML Algorithm to Improve Radar Detection of Golf Balls Using MFCW Radar"]}]}],"canonical_facts":{"dc:contributor.advisor":["Grootboom, L. L.","Steyn, W.","Niesler, T."],"dc:contributor.other":["Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering."],"dc:creator":["Holtshausen, Damian Luke"],"dc:date.accessioned":["2026-04-21T13:02:42Z"],"dc:date.available":["2026-04-21T13:02:42Z"],"dc:date.issued":["2026-03"],"dc:description":["Thesis (MEng)--Stellenbosch University, 2026.","Holtshausen, D. L. 2026. Development of a ML Algorithm to Improve Radar Detection of Golf Balls Using MFCW Radar. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/4937e1fc-e3d8-4bf4-9cfe-98ea33d8af2c"],"dc:description.abstract":["This thesis addresses the challenge of reliably detecting small targets, specifically golf balls, using Multi-Frequency Continuous-Wave (MFCW) radar in environments contaminated by significant noise and clutter. Although radar systems are effective at tracking targets, the performance of a radar system is heavily constrained by unwanted returns from external sources, such as fans, which often generate Doppler frequencies comparable to the target return signal. A comprehensive MFCW radar system simulation was developed, incorporating realistic physics and detailed models of critical interference targets. This enabled the design and testing of various detection algorithms. This thesis compares Constant False Alarm Rate (CFAR) detectors (including CA-, OS-, VI-, TM-, CHA- and WAI-CFAR) against modern Machine Learning (ML) techniques, including multinomial and binary Logistic Regression (LR) and Gradient Boosted Tree classifiers (XGBoost). For the ML models, a bin specific classification approach was explored, which utilised features which were engineered from statistical properties and signal peak characteristics. The models were trained on data collected from four international golf driving ranges. Results confirmed that classical CFAR detectors struggled in this cluttered environment, as seen by the false positive rates (FPR) which far exceeded the threshold that each detector was designed for. In contrast, the ML techniques, specifically XGBoost, which was used in a bin specific classifier, performed significantly better. The best model obtained a Recall (Probability of Detection) of 88.84%, with a False Positive Rate (FPR) of 0.431%. While the best performing CFAR detector, OS-CFAR obtained a recall of 81.42% with an FPR of 0.635%. These results indicate that ML models are capable of learning complex features beyond what is capable with CFAR detectors, allowing for more reliable detection of targets in environments with a low Signal to Clutter Ratio (SCR)."],"dc:identifier.uri":["https://scholar.sun.ac.za/handle/10019.1/136067"],"dc:language.iso":["en"],"dc:publisher":["Stellenbosch : Stellenbosch University"],"dc:title":["Development of a ML Algorithm to Improve Radar Detection of Golf Balls Using MFCW Radar"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T04:40:09Z"}