AI Chat Paper
Note: Please note that the following content is generated by AMiner AI. SciOpen does not take any responsibility related to this content.
{{lang === 'zh_CN' ? '文章概述' : 'Summary'}}
{{lang === 'en_US' ? '中' : 'Eng'}}
Chat more with AI
PDF (1.6 MB)
Collect
Submit Manuscript AI Chat Paper
Show Outline
Outline
Show full outline
Hide outline
Outline
Show full outline
Hide outline
Publishing Language: Chinese | Open Access

Optimized object detection and dataset generation for waste classification in real-world scenarios

Jianfeng Yang, Hong Yan, Tianle Mei, Ping Wang, Rui Cao( )
College of Computer Science, Northwest University, Xi’an 710127, China
Show Author Information

Abstract

To address challenges of complex backgrounds and small targets in real-world waste classification, this study proposes an enhanced object detection framework integrating synthetic data generation with model architecture optimization. The generated trash classification dataset with 12 Categories (GTC12) was constructed by combining semantic descriptions from large language models and images synthesized via Stable Diffusion, employing a filtering mechanism to ensure high quality. Compared with existing public benchmarks, the synthetic dataset exhibits closer alignment with real scenes in color complexity and object-to-image area ratio. For model design, a dynamic receptive-field convolution attention module (RFAC) and a nonlinear mapping module were introduced based on Vision Transformer Adapter (ViT-Adapter) to strengthen feature extraction and robustness against complex backgrounds. On the hybrid "real+synthetic" dataset GTC12-Mix, the enhanced model’s mean average precision (mAP50:95) increased from 0.77 to 0.80; accuracy increased from 0.85 to 0.86 in fine-grained binary classification experiments on the class-balanced dataset GTC12-Bal. Results demonstrate high diversity and realism of the synthetic dataset, along with superior detection capability of the proposed model for complex backgrounds and small targets.

CLC number: TP391.41

References

【1】
【1】
 
 
Journal of Northwest University (Natural Science Edition)
Pages 1136-1149

{{item.num}}

Comments on this article

Go to comment

< Back to all reports

Review Status: {{reviewData.commendedNum}} Commended , {{reviewData.revisionRequiredNum}} Revision Required , {{reviewData.notCommendedNum}} Not Commended Under Peer Review

Review Comment

Close
Close
Cite this article:
Yang J, Yan H, Mei T, et al. Optimized object detection and dataset generation for waste classification in real-world scenarios. Journal of Northwest University (Natural Science Edition), 2026, 56(5): 1136-1149. https://doi.org/10.16152/j.cnki.xdxbzr.2026-05-014

0

Views

0

Downloads

0

Crossref

0

CSCD

Received: 10 February 2026
Revised: 13 March 2026
Published: 25 October 2026
© The Editorial Department of Journal of Northwest University(Natural Science Edition)2026.

This is an open access article under the CC BY-NC-ND 4.0 license (https://creativecommons.org/licenses/by-nc-nd/4.0/).