DSResECA: A Lightweight CNN with Efficient Channel Attention for Organic and Recyclable Waste Image Classification

Zamzami, Amstrong Roosevelt (2026) DSResECA: A Lightweight CNN with Efficient Channel Attention for Organic and Recyclable Waste Image Classification. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211191-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Waste separation plays an important role in supporting sustainable environmental practices; however, in many areas, the sorting process is still carried out manually and often inconsistently. This study presents an image-based approach for identifying organic and recyclable waste by developing a Convolutional Neural Network (CNN) architecture called DSResECA. The model combines Depthwise-Separable Convolutions to reduce computational load, residual connections to stabilize deeper feature learning, and an Efficient Channel Attention mechanism that enables the network to focus on more informative and relevant visual cues. A publicly available waste image dataset was used in this study, and several data augmentation techniques, including MixUp, were applied to improve result consistency. To make the system practical for everyday use, the trained model was deployed within a web application where users can upload an image or capture one directly from their device’s camera. The system then provides an immediate classification result along with its confidence score.
================================================================================================================================
Pemilahan sampah memiliki peran penting dalam mendukung praktik lingkungan yang berkelanjutan, namun di banyak daerah proses pemilahan masih dilakukan secara manual dan sering kali tidak konsisten. Penelitian ini menghadirkan pendekatan berbasis citra untuk mengidentifikasi sampah organik dan daur ulang dengan mengembangkan arsitektur Convolutional Neural Network (CNN) yang disebut DSResECA. Model ini menggabungkan Depthwise-Separable Convolutions untuk mengurangi beban komputasi, residual connections untuk membantu menstabilkan pembelajaran fitur yang lebih dalam, serta mekanisme Efficient Channel Attention yang membuat jaringan lebih fokus pada petunjuk visual yang lebih informatif dan relevan. Dataset gambar sampah yang tersedia secara publik digunakan dalam penelitian ini, dan beberapa teknik augmentasi data, termasuk MixUp, diterapkan untuk meningkatkan konsistensi hasil. Agar sistem ini dapat digunakan secara praktis dalam kehidupan sehari-hari, model yang telah dilatih diterapkan ke dalam sebuah aplikasi web, di mana pengguna dapat mengunggah gambar atau mengambil foto langsung dari kamera perangkat mereka. Sistem kemudian memberikan hasil klasifikasi secara langsung beserta skor kepercayaannya.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, Waste Classification, Convolutional Neural Network, DSResECA, Image Recognition, Efficient Channel Attention, Web-Based Application
Subjects:	T Technology > T Technology (General) > T11 Technical writing. Scientific Writing T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Amstrong Roosevelt Zamzami
Date Deposited:	10 Jun 2026 01:19
Last Modified:	10 Jun 2026 01:19
URI:	http://repository.its.ac.id/id/eprint/133642

Actions (login required)

View Item