Amaludin, Moch Ahsan (2025) Sistem Sortir Deteksi Kerusakan Kardus Menggunakan Metode Convolution Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 2040211013-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
2040211013-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy |
Abstract
Sistem sortir konvensional yang berbasis sensor fisik seperti sensor ultrasonik dan proximity memiliki keterbatasan dalam mendeteksi cacat visual pada permukaan kardus, seperti sobekan, lubang, dan penyok. Hal ini menyebabkan ketidakefisienan dalam proses inspeksi produk pada jalur produksi. Penelitian ini bertujuan mengembangkan sistem sortir otomatis berbasis kamera dan deep learning dengan metode Convolutional Neural Network (CNN) dan arsitektur YOLOv8 untuk mendeteksi kerusakan kardus secara real time. Proses pengembangan sistem meliputi akuisisi dataset citra kardus, pelabelan menggunakan Roboflow, pelatihan model menggunakan google colab, dan implementasi sistem dengan integrasi perangkat keras seperti ESP32, solenoid pneumatik, dan kamera pengawas. Evaluasi dilakukan menggunakan berbagai metrik seperti confusion matrix, precision, recall, F1-score, mAP, dan akurasi. Hasil pengujian menunjukkan bahwa model YOLOv8 mampu mendeteksi kardus rusak dengan nilai recall 99,01%, precision 95,26%, F1-score 97,10%, dan average precision 94,32%, serta memberikan respons aktuasi di bawah 1 detik saat diuji secara real time. Sistem ini terbukti efektif dan responsif dalam melakukan klasifikasi visual objek serta mendukung penyortiran otomatis pada skala prototipe. Dengan demikian, penerapan CNN dan YOLO dalam penelitian ini memberikan solusi teknologi yang presisi dan adaptif untuk kebutuhan industri manufaktur modern.
===============================================================================================================================
Conventional sorting systems based on physical sensors such as ultrasonic and proximity sensors have limitations in detecting visual defects on cardboard surfaces, such as tears, holes, and dents. This causes inefficiency in the product inspection process on the production line. This research aims to develop an automatic sorting system based on cameras and deep learning using the Convolutional Neural Network (CNN) method and the YOLOv8 architecture to detect cardboard damage in real time. The system development process includes acquiring a cardboard image dataset, labeling using Roboflow, training the model using Google Colab, and implementing the system with hardware integration such as ESP32, pneumatic solenoids, and surveillance cameras. Evaluation was performed using various metrics such as confusion matrix, precision, recall, F1-score, mAP, and accuracy. Test results showed that the YOLOv8 model was able to detect damaged cardboard with a recall of 99.01%, precision of 95.26%, F1-score of 97.10%, and average precision of 94.32%, and provided an actuation response of less than 1 second when tested in real time. This system has proven to be effective and responsive in performing visual object classification and supports automatic sorting on a prototype scale. Thus, the application of CNN and YOLO in this study provides a precise and adaptive technological solution for the needs of modern manufacturing industries.
Actions (login required)
 |
View Item |