PENINGKATAN AKURASI, PRESISI DAN SENSITIFITAS FACE DETECTION MULTI CLASS PADA ALAT PELINDUNG DIRI DENGAN METODE ALGORITMA YOLOv7

DINATA, ILHAM WIBAWA KUSUMAH (2024) PENINGKATAN AKURASI, PRESISI DAN SENSITIFITAS FACE DETECTION MULTI CLASS PADA ALAT PELINDUNG DIRI DENGAN METODE ALGORITMA YOLOv7. S2 thesis, Universitas Mercu Buana - Menteng.

[img] Text (Cover)
55422110002-Ilham Wibawa Kusumah Dinata-01 Cover - Ilham Wibawa.pdf
Download (977kB)
[img] Text (Abstrak)
55422110002-Ilham Wibawa Kusumah Dinata-02 Abstrak - Ilham Wibawa.pdf
Download (390kB)
[img] Text (Bab 1)
55422110002-Ilham Wibawa Kusumah Dinata-03 Bab 1 - Ilham Wibawa.pdf
Restricted to Registered users only
Download (522kB)
[img] Text (Bab 2)
55422110002-Ilham Wibawa Kusumah Dinata-04 Bab 2 - Ilham Wibawa.pdf
Restricted to Registered users only
Download (1MB)
[img] Text (Bab 3)
55422110002-Ilham Wibawa Kusumah Dinata-05 Bab 3 - Ilham Wibawa.pdf
Restricted to Registered users only
Download (2MB)
[img] Text (Bab 4)
55422110002-Ilham Wibawa Kusumah Dinata-06 Bab 4 - Ilham Wibawa.pdf
Restricted to Registered users only
Download (2MB)
[img] Text (Bab 5)
55422110002-Ilham Wibawa Kusumah Dinata-07 Bab 5 - Ilham Wibawa.pdf
Restricted to Registered users only
Download (427kB)
[img] Text (Daftar Pustaka)
55422110002-Ilham Wibawa Kusumah Dinata-08 Daftar Pustaka - Ilham Wibawa.pdf
Restricted to Registered users only
Download (392kB)
[img] Text (Lampiran)
55422110002-Ilham Wibawa Kusumah Dinata-09 Lampiran - Ilham Wibawa.pdf
Restricted to Registered users only
Download (2MB)
[img] Text (Lembar Keabsahan)
55422110002-Ilham Wibawa Kusumah Dinata- Hasil Scan Formulir Pernyataan Keabsahan dan Persetujuan Publikasi Tugas Akhir - Ilham Wibawa.pdf
Restricted to Repository staff only
Download (141kB)

Abstract

Dalam dunia kerja di bidang konstruksi alat pelindungan diri merupakan hal yang sangat vital bagi pekerja untuk melindungi diri dari kecelakaan kerja sehingga pekerja dapat melakukan pekerjaannya secara aman guna meningkatkan hasil kerja dan produktivitas kerja. Face detection adalah langkah kunci dalam penerapan sistem pengenalan wajah, namun kemampuan komputasi perangkat tersebut relatif terbatas. Solusi terbaik untuk masalah dengan menggunakan algoritma YOLOv7 yang lebih cepat dan efisien. Sistem face detection alat pelindung diri menjadi alternatif untuk meminimalisir kelalaian pekerja dalam pemakaian perlengkapan wajib dalam bekerja. Metode yang digunakan dalam penelitian ini adalah YOLO (You Only Look Once). Penelitan ini membahas tentang implementasi model algoritma YOLOv7 dalam peningkatan akurasi, presisi dan sensitifitas untuk face detection alat pelindung diri. Objek yang akan di deteksi adalah wajah terhadap pemakaian dan tanpa alat pelindung diri yaitu helm pengaman, kacamata pengaman serta masker. Pada penelitian sebelumnya telah melakukan penelitian face detection penggunaan masker dengan metode yang sama menggunakan YOLOv7 memperoleh akurasi deteksi signifikan untuk semua kelas, memakai masker, bukan memakai masker dan pemakaian masker yang salah dengan akurasi tertinggi 92,2%. Hasil penelitian yang peneliti lakukan memperoleh nilai akurasi saat pengujian yang sangat baik yaitu 99,72% dengan objek multi class pada batch 16 dan epoch 500, sehingga diperoleh peningkatan akurasi sebesar 7,52% dari penelitian sebelumnya. Tidak hanya itu peneliti melakukan pengujian experimental dengan menggunakan dataset yang sama menggunakan algoritma YOLOv5, diperoleh nilai tertinggi sebesar 95,01% pada model YOLOv5x batch 16 epoch 300 sehingga diperoleh peningkatan akurasi sebesar 4,71%. Selain itu, algoritma YOLOv7 memperoleh nilai tertinggi untuk presisi sebesar 99,6% dan sensitifitas sebesar 100% dengan terdeteksinya semua kelas objek dibandingkan dengan algoritma YOLOv5, objek kelas tanpa masker tidak dapat terdeteksi. Algoritma YOLOv7 membutuhkan perangkat keras beberapa kali lebih murah dan dapat dilatih lebih cepat pada kumpulan data kecil tanpa bobot yang telah dilatih sebelumnya daripada algoritma YOLOv5. In the world of work in the field of construction, personal protective equipment is very vital for workers to protect themselves from work accidents so that workers can do their work safely to increase work output and work productivity. Face detection is a key step in the implementation of facial recognition systems, but the computing capabilities of these devices are relatively limited. The best solution to the problem is to use the faster and more efficient YOLOv7 algorithm. The face detection system of personal protective equipment is an alternative to minimize worker negligence in using mandatory equipment at work. The method used in this study is YOLO (You Only Look Once). This research discusses the implementation of the YOLOv7 algorithm model in improving accuracy, precision and sensitivity for face detection of personal protective equipment. The object to be detected is the face to wear and without personal protective equipment, namely safety helmets, safety glasses and masks. In the previous study, a face detection study on the use of masks with the same method using YOLOv7 obtained significant detection accuracy for all classes, wearing masks, not wearing masks, and wearing the wrong mask with the highest accuracy of 92.2%. The results of the research conducted by the researcher obtained an excellent accuracy value during testing, which was 99.72% with multi-class objects in batch 16 and epoch 500, so that an increase in accuracy of 7.52% was obtained from the previous study. Not only that the researchers conducted an experimental test using the same dataset using the YOLOv5 algorithm, obtained the highest score of 95.01% in the YOLOv5x batch 16 epoch 300 model so that an increase in accuracy of 4.71% was obtained. In addition, the YOLOv7 algorithm obtained the highest score for precision of 99.6% and sensitivity of 100% with the detection of all object classes compared to the YOLOv5 algorithm, which unmasked class objects cannot be detected. The YOLOv7 algorithm requires several times less hardware and can be trained faster on small, pre-trained weightless datasets than the YOLOv5 algorithm.highest score of 95.01% in the YOLOv5x batch 16 epoch 300 model so that an increase in accuracy of 4.71% was obtained. In addition, the YOLOv7 algorithm obtained the highest score for precision of 99.6% and sensitivity of 100% with the detection of all object classes compared to the YOLOv5 algorithm, which unmasked class objects cannot be detected. The YOLOv7 algorithm requires several times less hardware and can be trained faster on small, pre-trained weightless datasets than the YOLOv5 algorithm.

Item Type: Thesis (S2)
NIM/NIDN Creators: 55422110002
Uncontrolled Keywords: Face detection, alat pelindung diri, algoritma YOLOv7. Face detection, personal protective equipment, YOLOv7 algorithm.
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: NAYLA AURA RAYANI
Date Deposited: 09 Aug 2024 05:05
Last Modified: 09 Aug 2024 05:05
URI: http://repository.mercubuana.ac.id/id/eprint/90099

Actions (login required)

View Item View Item
UMB Repository is powered by EPrints 3 which is developed by the School of Electronics and Computer Science at the University of Southampton. More information and software credits.