Show simple item record

dc.contributor.advisorEscobedo Cárdenas, Edwin Jonathan
dc.contributor.authorCardenas Rondoño, Jonathan Bruce
dc.contributor.authorVasquez Espinoza, Ners Armando
dc.date.accessioned2026-01-26T20:43:07Z
dc.date.available2026-01-26T20:43:07Z
dc.date.issued2025
dc.identifier.urihttps://hdl.handle.net/20.500.12724/24294
dc.description.abstractLa metodología implementada comprendió etapas esenciales como el preprocesamiento de datos, la configuración e implementación de los modelos y el entrenamiento mediante estrategias de aprendizaje por transferencia. La validación se llevó a cabo en un entorno de video simulado, seguida de un análisis comparativo de las métricas obtenidaEl desempeño de los modelos fue evaluado empleando indicadores como la precisión promedio (mAP), el recall y el tiempo de inferencia. Los resultados mostraron que YOLO v8 alcanzó un mAP50 de 0.921, un recall de 0.829 y un tiempo de inferencia de 14.1 milisegundos. En contraste, YOLO-NAS obtuvo un mAP50 de 0.813, un recall de 0.903 y un tiempo de 17.8 milisegundos. Finalmente, RT-DETR registró un mAP de 0.887, un recall de 0.819 y un tiempo de 15.9 milisegundos. Aunque RT-DETR no logró el mAP más elevado, demostró un desempeño más estable en condiciones submarinas simuladas, lo cual evidencia su potencial en sistemas de monitoreo ambiental automatizadoes_PE
dc.description.abstractThis research addresses the global challenge of marine pollution, with particular emphasis on that generated by plastic waste. It applies real-time object detection methods powered by deep learning algorithms to identify plastic debris in underwater settings. A comparative analysis was performed among YOLO v8, YOLO-NAS, and RT-DETR models to assess their detection performance.The methodology comprised essential phases including data preprocessing, model configuration, and training using transfer learning techniques. Evaluation was conducted within a simulated video environment, followed by a comparative interpretation of the obtained outcomes.Model efficiency was measured through key indicators such as mean average precision (mAP), recall, and inference time. The YOLO v8 model achieved a mAP50 of 0.921, a recall of 0.829, and an inference time of 14.1 milliseconds. Conversely, YOLO-NAS obtained a mAP50 of 0.813, a recall of 0.903, and an inference time of 17.8 milliseconds. Lastly, RT-DETR reached a mAP of 0.887, a recall of 0.819, and an inference time of 15.9 milliseconds.Although RT-DETR did not record the highest mAP, it exhibited more consistent behavior in simulated underwater conditions, underlining its suitability for automated environmental monitoring systems.en_EN
dc.formatapplication/pdf
dc.language.isoeng
dc.publisherUniversidad de Lima
dc.rightshttps://purl.org/coar/access_right/c_abf2
dc.rights.urihttps://creativecommons.org/licenses/by-nc-sa/4.0/*
dc.subjectPendientees_PE
dc.titleUnderwater plastic waste detection with yolo and vision transformer modelsen_EN
dc.typehttps://purl.org/coar/resource_type/c_7a1f
thesis.degree.levelTitulo profesional
thesis.degree.disciplineIngeniería de Sistemas
thesis.degree.grantorUniversidad de Lima. Facultad de Ingeniería
dc.publisher.countryPE
dc.type.otherTesis
thesis.degree.nameIngeniero de Sistemas
renati.advisor.orcidhttps://orcid.org/0000-0003-2034-513X
renati.discipline612076
dc.identifier.isni0000000121541816
renati.author.dni71483402
renati.author.dni76964482
renati.levelhttps://purl.org/pe-repo/renati/level#tituloProfesional
renati.advisor.dni45211755
renati.jurorMayhua Quispe, Angela Gabriela
renati.jurorEscobedo Cárdenas, Edwin Jonathan
renati.jurorTincopa Flores, Jean Pierre
renati.typehttps://purl.org/pe-repo/renati/type#tesis
dc.subject.ocdehttps://purl.org/pe-repo/ocde/ford#2.11.04
ulima.catOI


Files in this item

Thumbnail
Thumbnail
Thumbnail

This item appears in the following Collection(s)

Show simple item record

https://purl.org/coar/access_right/c_abf2
Except where otherwise noted, this item's license is described as https://purl.org/coar/access_right/c_abf2