International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 04 | Apr 2025
p-ISSN: 2395-0072
www.irjet.net
REAL-TIME ORGANIC AND INORGANIC OBJECT DETECTION USING YOLO MODEL Ashlin Suba J, Jeya Shree R, Ragul Vignesh M, Mohana Priya G Dept. Computer Science and Business Systems,Nehru Institute of Engineering and Technology,Coimbatore, TamilNadu,India, Dept. Computer Science and Business Systems,Nehru Institute of Engineering and Technology,Coimbatore, TamilNadu,India, Assistant Professor(SG), Dept. Computer Science and Business Systems,Nehru Institute of Engineering and Technology,Coimbatore,TamilNadu,India, Dept. Computer Science and Business Systems,Nehru Institute of Engineering and Technology,Coimbatore, TamilNadu,India, ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Efficient and accurate classification of organic
systems that can significantly improve segregation accuracy and speed. Among these, the YOLO (You Only Look Once) family of algorithms is recognized for its superior real-time object detection performance. This study investigates the application of the latest YOLOv8 model to accurately detect and classify various waste materials, aiming to enhance the operational efficiency and scalability of modern waste management systems.
and inorganic materials is critical for advancing smart waste management, recycling operations, environmental monitoring, and automated sorting systems. Conventional manual classification approaches are often inefficient, labor-intensive, and susceptible to human error. This study proposes a realtime object detection framework utilizing the YOLOv8 (You Only Look Once, version 8) deep learning model to automate the identification and classification of diverse waste materials. The model was trained on a custom-curated dataset encompassing common organic items (such as fruits, vegetables, and paper) and inorganic items (including plastics, metals, and glass). Experimental results demonstrate that the proposed system achieves high classification accuracy and robust performance under real-world conditions. Moreover, it outperforms traditional object detection and image processing techniques in terms of inference speed and classification precision. These findings underscore the potential of the YOLOv8-based framework for scalable, real-time deployment in intelligent waste management and recycling infrastructures.
2. METHODOLOGY 2.1. Data Collection and Preprocessing A comprehensive dataset was curated, consisting of thousands of images representing a wide variety of organic and inorganic waste materials captured under diverse environmental conditions. Each image was meticulously annotated with bounding boxes and corresponding class labels to facilitate supervised learning. To improve the model’s robustness and generalization, various data augmentation techniques were applied, including rotation, scaling, and color adjustments. 2.2. Model Architecture The YOLOv8 model was chosen for this study due to its enhanced accuracy and inference speed compared to earlier versions. YOLOv8 incorporates an anchor-free detection mechanism and employs a more efficient backbone network, enabling real-time object detection with high precision. Transfer learning was leveraged by initializing the model with pre-trained weights, followed by fine-tuning on the custom waste classification dataset to adapt the model to the specific task.
Key Words: YOLO, deep learning, object detection, realtime classification, organic materials, inorganic materials, custom dataset, waste management.
1.INTRODUCTION The growing global challenges associated with waste management demand innovative and efficient solutions for the segregation of organic and inorganic materials. Organic waste, which is biodegradable, includes items such as food scraps and paper, while inorganic waste consists of nonbiodegradable materials like plastics and metals. Traditional manual segregation methods are labor-intensive, timeconsuming, and prone to human error, often resulting in reduced efficiency and effectiveness in recycling processes.
2.3. Training and Hyperparameter Tuning Model training was performed using the Adam optimizer, coupled with a learning rate scheduler to balance convergence speed and training stability. Critical hyperparameters such as batch size, input image resolution, number of epochs, and learning rate were carefully tuned through iterative experimentation to maximize model performance. Throughout training, monitoring techniques
Recent advances in computer vision and deep learning have enabled the development of automated waste classification
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 1968