Claim Missing Document
Check
Articles

Found 3 Documents
Search
Journal : chain journal of computer technology computer engineering and informatics

Sistem Deteksi Penyakit Gigi Berbasis Deep Learning Menggunakan YOLOv8 dan ResNet-18 Bagas Aditya; Rully Pramudita
CHAIN: Journal of Computer Technology, Computer Engineering, and Informatics Vol. 4 No. 3 (2026): Volume 4 Number 3 July 2026
Publisher : PT. Tech Cart Press

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58602/chain.v4i3.280

Abstract

Penyakit gigi dan mulut menjadi salah satu masalah kesehatan yang paling umum di Indonesia, dengan prevalensi mencapai 57,6% populasi menurut Survei Kesehatan Indonesia. Proses diagnosis konvensional masih sangat bergantung pada pemeriksaan visual manual oleh dokter gigi, yang memiliki keterbatasan dari segi waktu, subjektivitas, dan aksesibilitas. Penelitian ini mengembangkan sistem deteksi penyakit gigi berbasis deep learning yang mengintegrasikan arsitektur YOLOv8 untuk object detection dan ResNet-18 untuk image classification dalam sebuah pipeline ensemble. Sistem dirancang untuk mendeteksi enam jenis kelainan gigi: karies, karang gigi, radang gusi, hipodontia, sariawan, dan diskolorasi gigi dari foto kamera ponsel. Dataset yang digunakan berjumlah 11.957 citra yang dibagi menjadi 70% data latih, 15% validasi, dan 15% pengujian. Teknik weighted sampling diimplementasikan untuk menangani ketimpangan kelas dengan rasio 7,96x. Pelatihan ResNet-18 menggunakan optimizer Adam (learning rate 0,001) dengan fungsi kerugian CrossEntropyLoss berbobot kelas dinamis. Hasil evaluasi menunjukkan YOLOv8 mencapai mAP@50 sebesar 88,17%, sementara ResNet-18 memperoleh akurasi klasifikasi 92,25% dengan F1-Score 92,37%. Validasi statistik 5-Fold Cross Validation mengonfirmasi stabilitas ResNet-18 (Standar Deviasi = ±0,45%) dan YOLOv8 (Standar Deviasi = ±1,95%). Sistem ini diimplementasikan dalam aplikasi web menggunakan FastAPI dan Next.js pada GPU NVIDIA T4, dengan latensi end-to-end 2-4 detik, serta dilengkapi modul Grad-CAM untuk interpretabilitas prediksi.
Sistem Monitoring dan Kontrol Larutan Nutrisi Hidroponik NFT Berbasis IoT Menggunakan EMA dengan Analisis Interferensi Sensor Rafif Zetta Rajendra Pragiwoko; Rully Pramudita
CHAIN: Journal of Computer Technology, Computer Engineering, and Informatics Vol. 4 No. 3 (2026): Volume 4 Number 3 July 2026
Publisher : PT. Tech Cart Press

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58602/chain.v4i3.291

Abstract

The Nutrient Film Technique (NFT) hydroponic system requires stable nutrient solution management, particularly pH and nutrient concentration parameters that affect plant nutrient absorption. Based on observations conducted at the Mega Regency greenhouse, pH and nutrient monitoring are still performed manually, causing changes in nutrient conditions to not be monitored continuously in real-time. This study aims to design and implement an Internet of Things (IoT)-based monitoring and control system for nutrient solutions in NFT hydroponics. The development method used is Rapid Application Development (RAD). The system was developed using an ESP32 microcontroller integrated with pH, TDS, DS18B20 temperature, and ultrasonic sensors, as well as actuators in the form of pH up, pH down, and AB Mix nutrient pumps. Sensor data were processed using the Exponential Moving Average (EMA) method to reduce reading noise, while the control process applied the hysteresis method. Monitoring data were sent to the ThingsBoard platform and displayed through a Cloud-based dashboard. The results showed that the system was capable of continuously monitoring pH, nutrient concentration, temperature, and water level, with sensor accuracy reaching 95.84% for pH, 98.66% for TDS, and 97.14% for ultrasonic sensors. The EMA method effectively reduced sensor reading fluctuations by up to 0.74 pH units, while the hysteresis method maintained parameters within the specified range through automatic pump activation. During integration, electrochemical interference was found between pH and TDS sensor probes in the same reservoir. Various solutions were attempted progressively, from voltage stabilization using capacitors, switching power mechanism via transistors, to probe relocation to different pipe flow points as the more effective final solution. The developed system supports hydroponic management more effectively and automatically
Timbangan Digital Berbasis AIOT Dengan Deteksi Otomatis Jenis Buah Menggunakan YOLOv8 Dan Infrastruktur VPS Tiana Ramdani; Rully Pramudita
CHAIN: Journal of Computer Technology, Computer Engineering, and Informatics Vol. 4 No. 3 (2026): Volume 4 Number 3 July 2026
Publisher : PT. Tech Cart Press

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58602/chain.v4i3.310

Abstract

PT Interskala Mandiri Indonesia relies on manual input of Price Look Up (PLU) codes on the keypad for digital weighing, which results in human errors and lower operational efficiency. This study presents the development of an AIoT-based digital scale that integrates YOLOv8 for automatic fruit classification and leverages a Virtual Private Server (VPS) as a centralized data management infrastructure. The ADDIE model is used as the research and development framework. The hardware is built using an ESP32 NodeMCU-32S microcontroller, an ESP32-S3 CAM for image capture, and a load cell with an HX711 module for precise weight measurement. The YOLOv8n model was trained on five fruit classes (fuji apple, orange, lemon, century pear, and dragon fruit) and deployed on a VPS backend via Flask API. Receipt printing is performed through a Bluetooth T3 thermal printer using RawBT software, while monitoring is conducted through a React.js dashboard. Test results show that YOLOv8n achieved mAP@50 of 99.5%, precision of 99.97%, recall of 100%, and F1-score of 100%. The load cell provided 99.74% accuracy with a 0.26% error tolerance. All 25 Black Box Testing scenarios returned a Successful status. Average end-to-end latency was 7.55 seconds. The system proved capable of eliminating manual PLU input, centralizing transaction management, and providing a digital scale modernization solution for the retail industry.