Klaisifikasi Jenis Mikrokontroler Menggunakan Convolutional Neural Network (CNN) dan Webcam Berbasis Raspberry Pi

Maharani, Ayu Dea Desty (2025) Klaisifikasi Jenis Mikrokontroler Menggunakan Convolutional Neural Network (CNN) dan Webcam Berbasis Raspberry Pi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2042221025-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Penelitian ini mengembangkan sistem klasifikasi jenis mikrokontroler berbasis citra RGB menggunakan Convolutional Neural Network (CNN) dengan tiga kelas objek, yaitu Arduino UNO, ESP32-S3, dan STM32, yang diimplementasikan pada Raspberry Pi. Akuisisi citra dilakukan menggunakan webcam JETE W7 di dalam chamber tertutup dengan variasi tegangan dimmer 60, 110, 165, dan 220 V untuk pengaturan intensitas cahaya. Seluruh citra diproses melalui tahapan cropping ROI, resize 128×128 piksel, normalisasi (rescale 1/255), dan augmentasi (rotasi, shift, zoom, serta brightness). Proses pelatihan CNN menghasilkan kurva pembelajaran yang stabil selama 33 epoch, dengan akurasi training meningkat dari 0,3690 menjadi 1,0000 dan presisi training dari 0,4286 menjadi 1,0000, sementara loss training menurun dari 1,1021 menjadi 0,0647. Pada validasi, akurasi dan presisi mencapai 1,0000 dan loss turun dari 0,9360 menjadi 0,0001781. Evaluasi pada dataset uji menunjukkan performa klasifikasi 100% untuk seluruh kelas, ditunjukkan oleh nilai precision = 1,00, recall = 1,00, dan F1-score = 1,00, serta confusion matrix dengan nilai TP = 60 per kelas, FP = 0, FN = 0, dan TN = 120 per kelas (total 180 citra uji). Sistem real-time pada Raspberry Pi mampu melakukan inferensi secara kontinu dan menghasilkan prediksi kelas berdasarkan probabilitas softmax; penerapan mekanisme threshold 0,49 digunakan sebagai batas keputusan untuk menolak prediksi saat nilai probabilitas maksimum berada di bawah ambang tersebut. Dengan demikian, sistem yang dirancang mampu mengklasifikasikan tiga jenis mikrokontroler secara otomatis dan mencapai kinerja evaluasi maksimum pada data uji terkontrol, sehingga menjawab rumusan masalah terkait perancangan sistem, pelatihan model, dan performa klasifikasi berbasis metrik evaluasi.
======================================================================================================================================
This study developed an RGB image-based classification system for microcontroller types using a Convolutional Neural Network (CNN) with three object classes—Arduino UNO, ESP32-S3, and STM32—implemented on a Raspberry Pi. Image acquisition was performed using a JETE W7 webcam inside an enclosed chamber, with light intensity regulated by varying dimmer voltages (60, 110, 165, and 220 V). All images underwent processing stages including Region of Interest (ROI) cropping, resizing to 128×128 pixels, normalization (rescaling by 1/255), and augmentation (rotation, shifting, zooming, and brightness adjustment). The CNN training process yielded stable learning curves over 33 epochs; training accuracy rose from 0.3690 to 1.0000 and training precision from 0.4286 to 1.0000, while training loss decreased from 1.1021 to 0.0647. During validation, accuracy and precision reached 1.0000, and loss dropped from 0.9360 to 0.0001781. Evaluation on the test dataset demonstrated 100% classification performance across all classes—indicated by precision, recall, and F1-score values ​​of 1.00—and a confusion matrix showing TP = 60 per class, FP = 0, FN = 0, and TN = 120 per class (based on a total of 180 test images). The real-time system on the Raspberry Pi performs continuous inference and generates class predictions based on softmax probabilities; a threshold mechanism set at 0.49 serves as the decision boundary to reject predictions when the maximum probability value falls below this limit. Thus, the designed system is capable of automatically classifying three types of microcontrollers and achieving maximum evaluation performance on controlled test data, thereby addressing the problem formulation regarding system design, model training, and classification performance based on evaluation metrics.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Klasifikasi citra, mikrokontroler, real-time, pengolahan citra, CNN, image classification, microcontroller, image processing
Subjects:	T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.62 Simulation T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7888.3 Digital computers
Divisions:	Faculty of Vocational > Instrumentation Engineering
Depositing User:	Ayu Dea Desty Maharani
Date Deposited:	31 Jul 2026 01:46
Last Modified:	31 Jul 2026 01:46
URI:	http://repository.its.ac.id/id/eprint/140861

Actions (login required)

View Item