Enhancing Fruit Fly Detection in Complex Backgrounds Using Transformer Architecture with Step Attention Mechanism

Lexin Zhang, Kuiheng Chen, Liping Zheng, Xuwei Liao, Feiyu Lu, Yilun Li, Yuzhuo Cui, Yaze Wu, Yihong Song () and Shuo Yan ()
Additional contact information
Lexin Zhang: China Agricultural University, Beijing 100083, China
Kuiheng Chen: China Agricultural University, Beijing 100083, China
Liping Zheng: China Agricultural University, Beijing 100083, China
Xuwei Liao: China Agricultural University, Beijing 100083, China
Feiyu Lu: China Agricultural University, Beijing 100083, China
Yilun Li: China Agricultural University, Beijing 100083, China
Yuzhuo Cui: China Agricultural University, Beijing 100083, China
Yaze Wu: China Agricultural University, Beijing 100083, China
Yihong Song: China Agricultural University, Beijing 100083, China
Shuo Yan: China Agricultural University, Beijing 100083, China

Agriculture, 2024, vol. 14, issue 3, 1-27

Abstract: This study introduces a novel high-accuracy fruit fly detection model based on the Transformer structure, specifically aimed at addressing the unique challenges in fruit fly detection such as identification of small targets and accurate localization against complex backgrounds. By integrating a step attention mechanism and a cross-loss function, this model significantly enhances the recognition and localization of fruit flies within complex backgrounds, particularly improving the model’s effectiveness in handling small-sized targets and its adaptability under varying environmental conditions. Experimental results demonstrate that the model achieves a precision of 0.96, a recall rate of 0.95, an accuracy of 0.95, and an F1-score of 0.95 on the fruit fly detection task, significantly outperforming leading object detection models such as YOLOv8 and DETR. Specifically, this research delves into and optimizes for challenges faced in fruit fly detection, such as recognition issues under significant light variation, small target size, and complex backgrounds. Through ablation experiments comparing different data augmentation techniques and model configurations, the critical contributions of the step attention mechanism and cross-loss function to enhancing model performance under these complex conditions are further validated. These achievements not only highlight the innovativeness and effectiveness of the proposed method, but also provide robust technical support for solving practical fruit fly detection problems in real-world applications, paving new paths for future research in object detection technology.

Keywords: fruit fly detection; deep learning in plants; transformer architecture; step attention mechanism; cross-loss function (search for similar items in EconPapers)
JEL-codes: Q1 Q10 Q11 Q12 Q13 Q14 Q15 Q16 Q17 Q18 (search for similar items in EconPapers)
Date: 2024
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
https://www.mdpi.com/2077-0472/14/3/490/pdf (application/pdf)
https://www.mdpi.com/2077-0472/14/3/490/ (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:gam:jagris:v:14:y:2024:i:3:p:490-:d:1358847

Access Statistics for this article

Agriculture is currently edited by Ms. Leda Xuan

More articles in Agriculture from MDPI
Bibliographic data for series maintained by MDPI Indexing Manager ().