Indrawan, Arie (2025) PENGEMBANGAN METODE YOU ONLY LOOK ONCE (YOLOv11n) DENGAN INTEGRASI TEKNIK SHADOW ELIMINTAION UNTUK DETEKSI OBJEK. S2 thesis, Universitas Mercu Buana-Menteng.
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (COVER)
55422120008-Arie Indrawan-01 Cover - arie indrawan.pdf Download (649kB) |
|
Text (BAB I)
55422120008-Arie Indrawan-02 Bab 1 - arie indrawan.pdf Restricted to Registered users only Download (142kB) |
|
Text (BAB II)
55422120008-Arie Indrawan-03 Bab 2 - arie indrawan.pdf Restricted to Registered users only Download (410kB) |
|
Text (BAB III)
55422120008-Arie Indrawan-04 Bab 3 - arie indrawan.pdf Restricted to Registered users only Download (566kB) |
|
Text (BAB IV)
55422120008-Arie Indrawan-05 Bab 4 - arie indrawan.pdf Restricted to Registered users only Download (1MB) |
|
Text (BAB IV)
55422120008-Arie Indrawan-06 Bab 5 - arie indrawan.pdf Restricted to Registered users only Download (138kB) |
|
Text (DAFTAR PUSTAKA)
55422120008-Arie Indrawan-08 Daftar Pustaka - arie indrawan.pdf Restricted to Registered users only Download (143kB) |
|
Text (LAMPIRAN)
55422120008-Arie Indrawan-09 Lampiran - arie indrawan.pdf Restricted to Registered users only Download (129kB) |
Abstract
Deteksi objek dapat membantu dalam menginterpretasikan sebuah citra yang menghasilkan informasi yang digunakan untuk pemantauan lalu lintas. Namun kinerja Deteksi objek seringkali terdegradasi oleh kondisi lingkungan, terutama keberadaan bayangan. Bayangan dapat mengaburkan fitur visual objek sehingga mempengaruhi akurasi dan robustness sistem deteksi. Penelitian ini bertujuan untuk mengevaluasi dampak penanganan bayangan pada model deteksi objek YOLOv11n dengan mengintegrasikan Shadow Elimination SpA-Former dan DC-ShadowNet ke dalam arsitektur model YOLOv11n. Dataset yang digunakan dalam penelitian ini terdiri dari 1547 gambar. Pengujian dilakukan dengan 11 skenario untuk menganalisis berbagai konfigurasi model. Skenario ini mencakup pengujian model YOLOv11n baseline pada dataset asli, model YOLOv11n yang dilatih dan diuji secara individual dengan dataset hasil shadow elimination dari SpA-Former dan DC-ShadowNet, serta model YOLOv11n yang dilatih dengan strategi kombinasi dataset asli dan dataset hasil shadow elimination. Hasil pengujian menunjukkan bahwa pelatihan model YOLOv11n secara individual pada dataset yang telah dipra-proses shadow elimination menurunkan kinerja dibandingkan dengan baseline. Namun, strategi kombinasi dataset terbukti secara signifikan meningkatkan kinerja model. Model YOLOv11n + Kombinasi Dataset DC-ShadowNet mencapai [email protected] tertinggi sebesar 0.909 dan [email protected]:0.95 tertinggi sebesar 0.628, serta Recall 0.874. Sementara itu, model YOLOv11n + Kombinasi Dataset SpA-Former menunjukkan Precision tertinggi sebesar 0.884 dan [email protected]:0.95 sebesar 0.616. penelitian ini memvalidasi bahwa pra-pemrosesan shadow elimination yang diintegrasikan melalui strategi kombinasi dataset, secara efektif meningkatkan robustness dan akurasi model deteksi objek dalam menghadapi tantangan bayangan. Object detection plays an important role in interpreting images to produce information that can be utilized for traffic monitoring. However, the performance of object detection often degrades under environmental conditions, particularly due to the presence of shadows. Shadows can obscure visual features of objects, thereby affecting the accuracy and robustness of detection systems. This study aims to evaluate the impact of shadow handling on the YOLOv11n object detection model by integrating Shadow Elimination using SpA-Former and DC-ShadowNet into the YOLOv11n architecture. The dataset used in this study consists of 1,547 images. Experiments were conducted across 11 scenarios to analyze various model configurations. These scenarios include the evaluation of the baseline YOLOv11n model on the original dataset, YOLOv11n models trained and tested individually with shadow-eliminated datasets generated by SpA-Former and DC-ShadowNet, as well as models trained with a combined strategy using both the original dataset and shadow-eliminated datasets. The results show that training YOLOv11n individually on shadow-eliminated datasets led to performance degradation compared to the baseline. However, the combined dataset strategy significantly improved model performance. The YOLOv11n + Combined Dataset DC-ShadowNet model achieved the highest [email protected] of 0.909 and the highest [email protected]:0.95 of 0.628, along with a Recall of 0.874. Meanwhile, the YOLOv11n + Combined Dataset SpA-Former model achieved the highest Precision of 0.884 and an [email protected]:0.95 of 0.616. This study validates that shadow elimination preprocessing, when integrated through a combined dataset strategy, effectively enhances the robustness and accuracy of object detection models in challenging shadow conditions.
| Item Type: | Thesis (S2) |
|---|---|
| NIM/NIDN Creators: | 55422120008 |
| Uncontrolled Keywords: | Deteksi objek, You Only Look Once (YOLO), Shadow Elimination, SpA-Former, DC-ShadowNet Object detection, You Only Look Once (YOLO), Shadow Elimination, SpA-Former, DC-ShadowNet. |
| Subjects: | 600 Technology/Teknologi > 620 Engineering and Applied Operations/Ilmu Teknik dan operasi Terapan > 621 Applied Physics/Fisika terapan |
| Divisions: | Pascasarjana > Magister Teknik Elektro |
| Depositing User: | NAIMAH NUR ISLAMIDIYANAH |
| Date Deposited: | 29 Aug 2025 03:34 |
| Last Modified: | 29 Aug 2025 03:34 |
| URI: | http://repository.mercubuana.ac.id/id/eprint/97242 |
Actions (login required)
 |
View Item |
