面向多目标医疗垃圾分类的智能识别分拣系统设计

PDF下载HTML

Title:

Design of an intelligent identification and sorting system used for classification of multiobjective medical waste

作者:

张歆羽^1,2, 杨钟亮¹, 周哲画¹, 张凇³, 毛新华⁴

1. 东华大学 机械工程学院, 上海 201620;
2. 青岛虚拟现实研究院有限公司, 山东 青岛 266100;
3. 曼彻斯特大学, 曼彻斯特 M13 9PL;
4. 北京中丽制机工程技术有限公司, 北京 101111

Author(s):

ZHANG Xinyu^1,2, YANG Zhongliang¹, ZHOU Zhehua¹, ZHANG Song³, MAO Xinhua⁴

1. School of Mechanical Engineering, Donghua University, Shanghai 201620, China;
2. Qingdao Virtual Reality Institute Co., Ltd., Qingdao 266100, China;
3. Manchester University, Manchester M13 9PL;
4. Beijing Chonglee Machinery Engineering Co., L

关键词:

机器视觉; 目标检测; Delta分拣系统; 机械设计; 人工智能; 医疗垃圾; 垃圾分类; 智能垃圾箱

Keywords:

machine vision; object detection; Delta sorting system; mechanical design; artificial intelligence; medical waste; garbage classification; intelligent dustbin

分类号:

TP241.3；TP391.41

DOI:

10.11992/tis.202204039

文献标志码:

2023-10-10

摘要:

医疗垃圾中存在大量的病毒和细菌，为解决医疗垃圾源头智能分类问题，开发了基于机器视觉和Delta机构的智能分拣平台样机，并提出一种三阶段的多目标医疗垃圾识别分拣（medical waste recognition- indexes-sorting，MWRIS）算法。第1阶段提出数据增强扩容的IE-YOLOv4算法建立起医疗垃圾识别模型，与Faster R-CNN、RetinaNet、CenterNet等5种模型比较；第2阶段索引分类模型用于管理分类规则；第3阶段定位分拣算法指导目标定位分拣。在集成了MWRIS算法的分拣样机上，采集14种，2 217张医疗样本图像，完成医疗垃圾分拣实验。结果表明，使用IE-YOLOv4的MWRIS算法对医疗垃圾识别准确率显著提升至99.30%，分拣实验对目标定位准确率达到96.17%，最终分类正确率为86.67%，验证了多目标医疗垃圾识别分拣系统的有效性。

Abstract:

Medical waste contains lots of viruses and bacteria. To intelligently sort medical waste from the source, an intelligent sorting platform based on machine vision and the Delta mechanism was developed, and a three-stage multiobjective recognition-indexes-sorting (MWRIS) algorithm was proposed. In the first stage, the IE-YOLOv4 algorithm of data enhancement and expansion was proposed to establish a medical waste identification model, which was compared with five models, including Faster R-CNN, RetinaNet, and CenterNet. In the second stage, the index classification model was used to manage the classification rules. In the third stage, the positioning sorting algorithm was used to guide target positioning and grabbing. For the sorting prototype integrated with the MWRIS algorithm, 2217 medical sample images of 14 kinds were collected, and the medical waste sorting experiment was completed. The results showed that the MWRIS algorithm using IE-YOLOv4 can significantly improve the accuracy of medical waste identification to 99.30%, the accuracy rate of target positioning in the sorting experiment reaches 96.17%, and the final classification accuracy reaches 86.67%, verifying the effectiveness of the proposed medical waste identification and sorting system.