Wibawa, Febrian Aji (2024) Sistem Identifikasi Aktivitas Merokok Berbasis Kamera Termal, Sensor Gas, dan Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.
Text
5022201228-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (8MB) | Request a copy |
Abstract
Aktivitas merokok, yang melibatkan pembakaran tembakau dan menghasilkan asap, telah menjadi masalah global yang merugikan kesehatan dan keamanan. Menurut Global Adult Tobacco Survey 2021, jumlah perokok dewasa di Indonesia meningkat sebanyak 8,8 juta orang dari 2011 hingga 2021. Peningkatan tersebut juga dapat meningkatkan dampak dari merokok seperti menjadi salah satu penyebab kebakaran. Oleh karena itu, pengembangan alat deteksi dan peringatan aktivitas merokok menjadi penting sebagai upaya pencegahan dan penanggulangan terhadap bahaya rokok. Selain itu, alat deteksi yang tersedia di pasaran umumnya hanya menggunakan satu jenis sensor berupa infrared, yang masih berpotensi menghasilkan false alarm dari uap air atau gas lainnya seperti pengharum ruangan. Hal ini menunjukkan bahwa sistem deteksi yang ada saat ini masih memiliki keterbatasan dalam hal akurasi dan reliabilitas. Teknologi inovatif berbasis AI yang lebih cepat, akurat, dan mudah diimplementasikan memiliki potensi untuk memberikan solusi efektif terhadap masalah global aktivitas merokok. Maka dari itu, konsep fusi sensor dianggap sangat penting dalam meningkatkan ketepatan dan keandalan deteksi pada penelitian ini. Penelitian ini menggunakan kamera thermal MLX90640 dan sensor gas MQ-2, MQ-7, dan MQ-135 untuk mendeteksi suhu nyala dan karakteristik gas dalam asap rokok. Selain itu, penelitian membandingkan tiga metode deep learning: CNN dengan kamera termal, LSTM dengan sensor gas, dan kombinasi keduanya (CNN-LSTM). Hasil pengujian realtime menunjukkan bahwa model CNN mencapai akurasi tertinggi 85% pada suhu ruangan 34⁰C dan 76.67% pada 29⁰C untuk penempatan alat di bawah, serta 100% untuk penempatan di atas dengan tinggi objek 'gangguan' ±150 cm. Model LSTM mencapai akurasi 82% untuk penempatan di bawah dan 95% untuk penempatan di atas. Gabungan CNNLSTM tidak menghasilkan peningkatan signifikan pada penempatan di bawah, namun mencapai akurasi 100% pada penempatan di atas dengan delay waktu 50 detik hingga 120 detik.
=================================================================================================================================
Smoking, which involves burning tobacco and producing smoke, has become a global problem that harms health and safety. According to the Global Adult Tobacco Survey 2021, the number of adult smokers in Indonesia increased by 8.8 million people from 2011 to 2021. This increase can also increase the impact of smoking such as being one of the causes of fires. Therefore, the development of smoking activity detection and warning devices is important as an effort to prevent and mitigate the dangers of smoking. In addition, detection devices available on the market generally only use one type of sensor in the form of infrared, which still has the potential to produce false alarms from water vapor or other gases such as air fresheners. This shows that current detection systems still have limitations in terms of accuracy and reliability. Innovative AI-based technologies that are faster, more accurate, and easier to implement have the potential to provide effective solutions to the global problem of smoking activity. Therefore, the concept of sensor fusion is considered very important in improving the accuracy and reliability of detection in this study. This research utilizes the MLX90640 thermal camera and MQ-2, MQ-7, and MQ-135 gas sensors to detect flame temperature and gas characteristics in cigarette smoke. In addition, the study compared three deep learning methods: CNN with thermal camera, LSTM with gas sensor, and a combination of both (CNN-LSTM). The realtime test results showed that the CNN model achieved the highest accuracy of 85% at room temperature of 34⁰C and 76.67% at 29⁰C for device placement below, as well as 100% for placement above with a 'nuisance' object height of ±150 cm. The LSTM model achieved 82% accuracy for bottom placement and 95% for top placement. The combined CNN-LSTM did not produce significant improvement in under-placement, but achieved 100% accuracy in overplacement with a time delay of 50 seconds to 120 seconds.
Item Type: | Thesis (Other) |
---|---|
Uncontrolled Keywords: | Aktivitas Merokok, CNN, LSTM, Kamera Termal, Sensor Gas, Smoking Activity, Thermal Camera, Gas Sensor |
Subjects: | T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems |
Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis |
Depositing User: | Febrian Aji Wibawa |
Date Deposited: | 30 Jul 2024 03:02 |
Last Modified: | 30 Jul 2024 03:02 |
URI: | http://repository.its.ac.id/id/eprint/109725 |
Actions (login required)
View Item |