Evander Tokang, - (2025) IMPLEMENTASI YOLOv11 DENGAN TEKNIK QUANTIZATION UNTUK DETEKSI IKLAN JUDI ONLINE DI INSTAGRAM. S1 thesis, Universitas Pendidikan Indonesia.
Abstract
Peredaran iklan judi online yang semakin masif di media sosial seperti Instagram menuntut solusi moderasi konten yang akurat dan efisien. Penelitian ini bertujuan untuk mengimplementasikan model deteksi objek YOLOv11 yang dioptimalkan menggunakan teknik post training quantization (PTQ) dalam tiga format numerik, yaitu FP32, FP16, dan INT8, untuk mendeteksi iklan judi online pada gambar statis. Dataset terdiri dari 500 gambar beranotasi, dilatih menggunakan varian YOLOv11n dan YOLOv11s, lalu dikonversi ke format FP32, FP16, dan INT8 untuk tiga platform eksekusi, yaitu TensorRT pada GPU, OpenVINO pada CPU, dan TensorFlow Lite pada perangkat Android. Evaluasi dilakukan berdasarkan akurasi model menggunakan metrik mean average precision (mAP), precision, recall, dan F1-score, serta efisiensi komputasi melalui pengukuran waktu inferensi, ukuran model, konsumsi RAM, CPU, dan daya baterai. Hasil menunjukkan bahwa model TensorRT INT8 mencatat waktu inferensi tercepat sebesar 3.5 ms dengan ukuran model hanya 6.64 MB, menjadikannya ideal untuk sistem server berbasis GPU. Model OpenVINO INT8 mencatat akurasi tertinggi dengan F1-score 99.9% dan mAP@0.5:.95 hingga 76.7%, sekaligus mempertahankan efisiensi tinggi pada sistem tanpa GPU. Sementara itu, model TFLite INT8 cocok untuk perangkat seluler karena ukurannya hanya 2.68 MB, konsumsi daya lebih hemat hingga 26.2%, dan recall sempurna sebesar 100%. Dengan demikian, teknik quantization terbukti efektif untuk mempertahankan akurasi sekaligus meningkatkan efisiensi lintas platform dalam penerapan sistem deteksi iklan judi online secara real-time. ----------- The widespread circulation of online gambling advertisements on social media platforms such as Instagram demands an accurate and efficient content moderation solution. This study aims to implement the YOLOv11 object detection model optimized using post training quantization (PTQ) in three numerical formats, namely FP32, FP16, and INT8, to detect online gambling ads in static images. The dataset consists of 500 annotated images, trained using the YOLOv11n and YOLOv11s variants, then converted into FP32, FP16, and INT8 formats for deployment on three execution platforms, namely TensorRT on GPU, OpenVINO on CPU, and TensorFlow Lite on Android devices. The evaluation was based on model accuracy using metrics such as mean average precision (mAP), precision, recall, and F1-score, as well as computational efficiency measured by inference time, model size, RAM usage, CPU load, and battery consumption. The results show that the TensorRT INT8 model achieved the fastest inference time of 3.5 ms with a model size of only 6.64 MB, making it ideal for GPU-based server systems. The OpenVINO INT8 model recorded the highest accuracy with an F1-score of 99.9% and mAP up to 76.7%, while maintaining high efficiency on systems without GPUs. Meanwhile, the TFLite INT8 model is suitable for mobile devices due to its size of only 2.68 MB, battery savings of up to 26.2%, and perfect recall of 100%. Thus, quantization has proven effective in maintaining accuracy while enhancing cross-platform efficiency for real-time detection of online gambling advertisements.
![[thumbnail of S_RPL_2102359_Title.pdf]](http://repository.upi.edu/style/images/fileicons/text.png "S_RPL_2102359_Title.pdf") |
Text
S_RPL_2102359_Title.pdf Download (1MB) |
|
Text
S_RPL_2102359_Chapter1.pdf Download (251kB) |
|
Text
S_RPL_2102359_Chapter2.pdf Restricted to Staf Perpustakaan Download (492kB) | Request a copy |
|
Text
S_RPL_2102359_Chapter3.pdf Download (571kB) |
|
Text
S_RPL_2102359_Chapter4.pdf Restricted to Staf Perpustakaan Download (1MB) | Request a copy |
|
Text
S_RPL_2102359_Chapter5.pdf Download (202kB) |
|
Text
S_RPL_2102359_Appendix.pdf Restricted to Staf Perpustakaan Download (455kB) | Request a copy |
| Item Type: | Thesis (S1) |
|---|---|
| Additional Information: | https://scholar.google.com/citations?hl=en&user=0cuKVK0AAAAJ ID SINTA Dosen Pembimbing: Mochamad Iqbal Ardimansyah: 6658552 Raditya Muhammad: 6682222 |
| Uncontrolled Keywords: | YOLOv11, quantization, iklan judi online, deteksi objek, moderasi konten, Instagram YOLOv11, quantization, online gambling ads, object detection, content moderation, Instagram |
| Subjects: | L Education > L Education (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General) |
| Divisions: | UPI Kampus cibiru > S1 Rekayasa Perangkaat Lunak |
| Depositing User: | Evander Tokang |
| Date Deposited: | 07 Aug 2025 03:15 |
| Last Modified: | 07 Aug 2025 03:15 |
| URI: | http://repository.upi.edu/id/eprint/135208 |
Actions (login required)
 |
View Item |