View in EDS

From Segmentation to Diagnosis: An End-to-End NMP-Mamba Framework and Expert Agent for Intelligent Peanut Root Nodule Analysis.

Saved in:
Bibliographic Details
Title: From Segmentation to Diagnosis: An End-to-End NMP-Mamba Framework and Expert Agent for Intelligent Peanut Root Nodule Analysis.
Authors: Hu, Yating (AUTHOR), Wang, Wei (AUTHOR), Zhang, Qi (AUTHOR), Li, Peiwu (AUTHOR)
Source: Journal of the ASABE. 2026, Vol. 69 Issue 3, p357-366. 10p.
Subjects: Image segmentation, Root-tubercles, Phenotypes, Precision farming, Nitrogen fixation, Deep learning
Abstract: Peanut nodule phenotyping provides an important basis for assessing symbiotic N fixation potential and guiding cultivation management. However, root images often exhibit high texture similarity between nodules and roots, with small-scale targets co-existing with dense occlusion clusters, which allows segmentation errors to accumulate along the counting pipeline and manifest as systematic bias. To address key bottlenecks, insufficient separability in homogeneous backgrounds, missed detections of small-scale nodules, and under-segmentation in dense scenes, this study proposed a Root Nodule Morphology-Prior Mamba Segmentation Network (NMP-Mamba). Specifically, Dual-Domain Spatial-Frequency Encoder (DSFE) was introduced to enhance discriminative representations between nodules and roots under texture-homogeneous conditions; Multi-scale Feature Aggregation Gating (MFAG) was incorporated to strengthen small-target responses and aggregate multi-scale features; and Geometry-Aware Distance-Field Regression (GADR) was employed to provide more stable splitting cues in dense occlusion regions. Experiments demonstrated that NMP-Mamba achieved consistent advantages in both segmentation and counting (Dice 95.35%, Nodule IoU 91.11%, MAE 5.33) while maintaining 51.6 FPS; ablation analyses further verified the complementary contributions of the components to counting-regression agreement and error-variance control. In addition, an end-cloud collaborative nodule diagnosis APP was built upon the model, with the predicted masks and phenotypic indicators packaged as structured inputs to an agriculture-oriented Expert Agent, enabling an end-to-end workflow from image segmentation and phenotypic quantification to diagnostic text and management recommendations and providing intuitive reference and assistance for nodule counting and phenotyping. [ABSTRACT FROM AUTHOR]
Copyright of Journal of the ASABE is the property of American Society of Agricultural & Biological Engineers and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Peanut nodule phenotyping provides an important basis for assessing symbiotic N fixation potential and guiding cultivation management. However, root images often exhibit high texture similarity between nodules and roots, with small-scale targets co-existing with dense occlusion clusters, which allows segmentation errors to accumulate along the counting pipeline and manifest as systematic bias. To address key bottlenecks, insufficient separability in homogeneous backgrounds, missed detections of small-scale nodules, and under-segmentation in dense scenes, this study proposed a Root Nodule Morphology-Prior Mamba Segmentation Network (NMP-Mamba). Specifically, Dual-Domain Spatial-Frequency Encoder (DSFE) was introduced to enhance discriminative representations between nodules and roots under texture-homogeneous conditions; Multi-scale Feature Aggregation Gating (MFAG) was incorporated to strengthen small-target responses and aggregate multi-scale features; and Geometry-Aware Distance-Field Regression (GADR) was employed to provide more stable splitting cues in dense occlusion regions. Experiments demonstrated that NMP-Mamba achieved consistent advantages in both segmentation and counting (Dice 95.35%, Nodule IoU 91.11%, MAE 5.33) while maintaining 51.6 FPS; ablation analyses further verified the complementary contributions of the components to counting-regression agreement and error-variance control. In addition, an end-cloud collaborative nodule diagnosis APP was built upon the model, with the predicted masks and phenotypic indicators packaged as structured inputs to an agriculture-oriented Expert Agent, enabling an end-to-end workflow from image segmentation and phenotypic quantification to diagnostic text and management recommendations and providing intuitive reference and assistance for nodule counting and phenotyping. [ABSTRACT FROM AUTHOR]
ISSN:27693295
DOI:10.13031/ja.16675