Smart segregation system: discrimination of degradable and non-degradable waste using deep learning towards environmental sustainability/
Ace Jester Alicante, Paulo R. Bendijo, Kenneth I. Lopena, Kaye Charm Chiara Malubay, and Maureen E. Ramirez.--
- Manila: Technological University of the Philippines, 2024.
- xii, 150pages: 29cm.
Bachelor's thesis
College of Industrial Technology.--
Includes bibliographic references and index.
Solid waste management continues to encounter significant challenges, particularly in waste segregation, where manual sorting is often inefficient and obsolete. With the aim of resolving this serious concern, an eco-friendly, low-cost, and automated waste sorting system was designed to detect, sort, and classify waste into four different categories: biodegradable, non-biodegradable, hazardous, and unsorted. The system integrates hardware components such as a camera module, servo motors, ultrasonic sensors, and a Raspberry Pi, combined with advance image processing techniques supported by machine learning and deep learning algorithms. Real-time system status is displayed on an LCD display, while bin capacity levels are visually and audibly indicated using LEDs and buzzers. The entire system was developed using an iterative prototyping approach, allowing for continuous improvement to enhance its performance. You Only Look Once version 8 deep learning model, also known as YOLOv8, was trained on a custom waste dataset for precise object detection and identification, with TensorFlow Lite seamlessly integrated to optimize efficient real-time processing. Smooth coordination was achieved between the camera module, servo motors, sorting mechanism, ultrasonic sensors, and the Raspberry Pi acting as the main controller. Extensive testing was conducted to assess classification accuracy, response time, system efficiency, processor performance, and detection reliability. Black-box testing confirmed the system functionality, while evaluated using ISO/IEC 25010 standards focusing on functionality, performance efficiency, usability, reliability, and maintainability through survey questionnaire with 10 waste management experts as respondents. The system achieved a user satisfaction score of 4.5 out of 5 on the Likert scale, which falls under the 'Very Good' category. This study establishes the vast potential of combining hardware with machine learning and deep learning algorithms to design innovative and environmentally friendly solutions for waste management automation.