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

Abstract

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 Region-Driven 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 Mamba-based 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 boundary-ambiguous 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.

Plant phenotyping relevance

植物病斑を画像から分割し、病害状態・重症度を推定する新規手法を開発し、公開データセットで検証しているため、植物フェノタイピング手法が中心である。

Code and data availability

The paper uses two public plant disease segmentation datasets as its phenotyping inputs, both explicitly linked in the data availability statement: the Leaf Disease Segmentation Dataset (Kaggle) and the PlantSeg dataset (Zenodo record 13762907). No author code or model release is mentioned.