An improved image segmentation method for pest with high-frequency information compensation
Smart Agricultural Technology · 2026-12 · 15 · 102566
DOI: 10.1016/j.atech.2026.102566 ↗Kurz gefasst
Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Yingqiong Peng und 5 weiteren Autor:innen. Die Arbeit nennt 69 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Natural Science Foundation of Jiangxi Province, National Natural Science Foundation of China, Education Department of Jiangxi Province.
Abstract (Originalsprache)
Pest detection is vital in precision agriculture, serving as a cornerstone for preserving plant health and implementing targeted pest management strategies. Among these pests, Drosophila poses a serious threat to fruits and vegetables, causing extensive crop damage and significant economic losses. Accurate segmentation of Drosophila is crucial for early detection and effective control, ensuring the protection of agricultural yields and sustainable plant health. Existing image segmentation methods often lose high-frequency information during downsampling processes, which negatively impacts the precision of Drosophila segmentation by discarding critical discriminative details and edges. To address this limitation, we propose a novel Drosophila segmentation approach named HFIC-DroSeg. Specifically, our method incorporates two key components: a High-Frequency Information Extraction Module, which adaptively extracts high-frequency information from images to capture richer details, and a Multi-Scale High-Frequency Aggregation Module. This module enhances the preservation of detailed information by aggregating high-frequency components across different scales and integrating edge branches to more accurately locate Drosophila edge information. Experimental results demonstrate that our proposed model achieves superior performance compared to existing methods. On our DrosophilaSet dataset, HFIC-DroSeg achieves the highest scores in Precision (0.93), Recall (0.94), Accuracy (0.95), mIoU (0.94), and Dice coefficient (0.96). Furthermore, the model demonstrates strong generalization capabilities on a crop insect pest dataset and performs well on the widely used PASCAL VOC 2012 segmentation benchmark.
Förderung
- Natural Science Foundation of Jiangxi Province
- National Natural Science Foundation of China
- Education Department of Jiangxi Province
Zitieren
APA
Peng, Y., Jin, Y., Li, Y., Liao, M., Xu, Y., & Yin, C. (2026). An improved image segmentation method for pest with high-frequency information compensation. Smart Agricultural Technology, 15, 102566. https://doi.org/10.1016/j.atech.2026.102566
Vancouver
Peng Y, Jin Y, Li Y, Liao M, Xu Y, Yin C. An improved image segmentation method for pest with high-frequency information compensation. Smart Agricultural Technology. 2026;15:102566. doi:10.1016/j.atech.2026.102566
BibTeX
@article{Peng2026,
author = {Peng, Yingqiong and Jin, Yuling and Li, Yueqian and Liao, Muxin and Xu, Yilu and Yin, Chengle},
title = {An improved image segmentation method for pest with high-frequency information compensation},
journal = {Smart Agricultural Technology},
year = {2026},
volume = {15},
pages = {102566},
doi = {10.1016/j.atech.2026.102566}
}
EUVIMED