Open Access

ARTICLE

An Improved YOLO-Based Waste Detection Model and Its Integration to Robotic Gripping Systems

Anjie Wang^1,2, Haining Jiao^1,2,*, Zhichao Chen^1,2,*, Jie Yang^1,2

1 School of Intelligent Manufacturing and Materials Engineering, Gannan University of Science and Technology, Ganzhou, 341000, China
2 School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, 341000, China

* Corresponding Authors: Haining Jiao. Email: ; Zhichao Chen. Email:

(This article belongs to the Special Issue: New Trends in Image Processing)

Computers, Materials & Continua 2025, 84(3), 5773-5790. https://doi.org/10.32604/cmc.2025.066852

Received 18 April 2025; Accepted 26 June 2025; Issue published 30 July 2025

Abstract

With the rapid development of the Internet of Things (IoT), artificial intelligence, and big data, waste-sorting systems must balance high accuracy, low latency, and resource efficiency. This paper presents an edge-friendly intelligent waste-sorting system that integrates a lightweight visual neural network, a pentagonal-trajectory robotic arm, and IoT connectivity to meet the requirements of real-time response and high accuracy. A lightweight object detection model, YOLO-WasNet (You Only Look Once for Waste Sorting Network), is proposed to optimize performance on edge devices. YOLO-WasNet adopts a lightweight backbone, applies Spatial Pyramid Pooling-Fast (SPPF) and Convolutional Block Attention Module (CBAM), and replaces traditional C3 modules (Cross Stage Partial Bottleneck with 3 convolutions) with efficient C2f blocks (Cross Stage Partial Bottleneck with 2 Convolutions fast) in the neck. Additionally, a Depthwise Parallel Triple-attention Convolution (DPT-Conv) operator is introduced to enhance feature extraction. Experiments on a custom dataset of nine waste categories conforming to Shanghai’s sorting standard (7,917 images) show that YOLO-WasNet achieves a mean average precision () of 96.8% and a precision of 96.9%, while reducing computational cost by 30% compared to YOLOv5s. On a Raspberry Pi 4B, inference time is reduced from 480 to 350 ms, ensuring real-time performance. This system offers a practical and viable solution for low-cost, efficient automated waste management in smart cities.

---

Keywords

---