Gradient-guided boundary-aware selective scanning with multi-scale context aggregation for plant lesion segmentation

Sep 29, 2026

Upload by Fahsai

Plant lesion segmentation targets pixel-level delineation of disease regions to enable early diagnosis, severity estimation, and precision agricultural interventions. The task is difficult because lesion scale varies from incipient micro-spots to merged areas and because lesion boundaries are often low-contrast and blend into healthy tissue. The GARDEN approach integrates multi-scale context aggregation to build scale-consistent lesion priors and a boundary-aware selective scanning mechanism guided by gradient-predicted boundaries for sharper, more reliable segmentation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Plant lesion segmentation challenges

Methods

GARDEN framework and MSCA module

Boundary-aware selective scanning with GGBP

Results

Performance on public plant disease datasets

Discussion

Boundary precision and scale robustness

TYPE Original Research PUBLISHED 23 December 2025 DOI 10.3389/fpls.2025.1727075 OPEN ACCESS EDITED BY Zhenghong Yu, Guangdong Polytechnic of Science and Technology, China REVIEWED BY Hao Lu, Huazhong University of Science and Technology, China Yang Long, Hubei Minzu University, China *CORRESPONDENCE Jia Guo [jia.guo.46@hosei.ac.jp](jia.guo.46@hosei.ac.jp)[ ](jia.guo.46@hosei.ac.jp)Junyi Xin [xinjunyi@hmc.edu.cn](xinjunyi@hmc.edu.cn) †These authors have contributed equally to this work and share ﬁrst authorship RECEIVED 17 October 2025 REVISED 26 November 2025 ACCEPTED 28 November 2025 PUBLISHED 23 December 2025 CITATION Sun G, Li T, Pan Y, Zhu Z, Yang T, Shao F, Guo J and Xin J (2025) Gradient-guided boundary-aware selective scanning with multi-scale context aggregation for plant lesion segmentation. Front. Plant Sci. 16:1727075 . doi: 10.3389/fpls.2025.1727075 COPYRIGHT © 2025 Sun, Li, Pan, Zhu, Yang, Shao, Guo and Xin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. Gradient-guided boundaryaware selective scanning with multi-scale context aggregation for plant lesion segmentation Guanqun Sun 1†, Tianshuo Li 1†, Yizhi Pan 1,2†, Zidan Zhu 1, Tianhua Yang 1, Feihe Shao 1, Jia Guo 3* and Junyi Xin 1* 1School of Information Engineering, Hangzhou Medical College, Hangzhou, Zhejiang, China, 2School of Information Science, Japan Advanced Institute of Science and Technology, Nomi, Japan, 3 Faculty of Computer and Information Sciences, Hosei University, Tokyo, Japan Introduction: Plant lesion segmentation aims to delineate disease regions at the pixel level to support early diagnosis, severity assessment, and targeted intervention in precision agriculture. However, the task remains challenging due to large variations in lesion scale—ranging from minute incipient spots to coalesced regions—and ambiguous, low-contrast boundaries that blend into healthy tissue. Methods: We present GARDEN, a Gradient-guided boundary-Aware RegionDriven Edge-reﬁNement network that uniﬁes multi-scale context modeling with selective long-range boundary reﬁnement. Our approach integrates a Multi-Scale Context Aggregation (MSCA) module to harvest contextual cues across diverse receptive ﬁelds, forming scale-consistent lesion priors to improve sensitivity to tiny lesions. Additionally, we introduce a Boundary-aware Selective Scanning (BASS) module conditioned on a Gradient-Guided Boundary Predictor (GGBP) . This module produces an explicit boundary prior to steer a Mambabased 2D selective scan, allocating long-range reasoning to boundary-uncertain pixels while relying on local evidence in conﬁdent interiors. Results: Validated across two public plant disease datasets, GARDEN achieves state-of-the-art results on both overlap and boundary metrics. Speciﬁcally, the model demonstrates pronounced gains on small lesions and boundaryambiguous cases . Qualitative results further show sharper contours and reduced spurious responses to illumination and viewpoint changes compared to existing methods. Discussion: By coupling scale robustness with boundary precision in a single architecture, GARDEN delivers accurate and reliable plant lesion segmentation. This method effectively addresses key challenges in the ﬁeld, offering a robust solution for automated disease analysis under challenging real-world conditions. KEYWORDS gradient-guided, Mamba, multi-scale context aggregation, plant lesion segmentation, selective scanning, state space models Frontiers in Plant Science 01 [frontiersin.org](frontiersin.org)

FAQs

What problem does plant lesion segmentation solve?

It labels disease regions in plant images at the pixel level, supporting early diagnosis, severity measurement, and targeted treatment in precision agriculture.

Why is the segmentation task challenging?

Lesions vary greatly in scale and appearance, and their boundaries often have low contrast that blends into healthy tissue.

What are the key components of the GARDEN method?

GARDEN combines Multi-Scale Context Aggregation (MSCA) for scale-consistent lesion priors with a boundary-aware selective scanning module guided by a Gradient-Guided Boundary Predictor (GGBP).

Sep 29, 2026

Upload by Fahsai