Morphology-guided detail enhancement and adaptive multiscale fusion for sugarcane stem node detection

Accurate localization of sugarcane stem nodes is essential for determining cutting positions and billet lengths in mechanized seed-cane cutting. However, detection remains difficult because the sugarcane stem nodes are small and densely distributed, while illumination variation and low-contrast boundaries further weaken their visual cues. Fine positional and boundary information is also prone to attenuation during successive downsampling in the YOLO11 backbone. To address these issues, this study proposes MPS-YOLO, an enhanced YOLO11n model for sugarcane stem node detection. A public sugarcane stem node dataset containing field and indoor images was used for model training and benchmark evaluation. First, a Node Morphology Detail Enhancement (NMDE) module is introduced into the shallow backbone to jointly extract horizontal, vertical, local-texture, and Sobel-gradient features, with channel-wise gating used to recalibrate fused representation. Second, the DySample operator is incorporated into the high-resolution feature fusion path to reconstruct fine spatial features through learned sampling offsets. Third, a high-resolution prediction branch is added to the neck to retain positional and boundary information for densely distributed sugarcane stem nodes. Under a fixed data split and unified training settings, the results averaged over five runs with different random seeds showed that MPS-YOLO improved precision, recall, [email protected], and [email protected]:0.95 over YOLO11n by 2.00, 2.86, 2.08, and 4.50 percentage points, respectively. The proposed model has approximately 2.7 M parameters and requires 11.8 GFLOPs. These results demonstrate improved detection and localization on the public benchmark and may support visual localization in mechanized seed-cane cutting.

Authors

Institutions

Publication Details

Journal
Frontiers in Plant Science
Published
2026-11-16
DOI
https://doi.org/10.3389/fpls.2026.2005858
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Morphology-guided detail enhancement and adaptive multiscale fusion for sugarcane stem node detection

Ning Li, Yong Bao, Lingmei Wu, Bineng Zhong et al.
5 links
Frontiers in Plant Science
Smart Agriculture and AI
article

Morphology-guided detail enhancement and adaptive multiscale fusion for sugarcane stem node detection

Ning Li, Yong Bao, Lingmei Wu, Bineng Zhong, Liqiang Zhang, Tao Cheng
article en

Abstract

Accurate localization of sugarcane stem nodes is essential for determining cutting positions and billet lengths in mechanized seed-cane cutting. However, detection remains difficult because the sugarcane stem nodes are small and densely distributed, while illumination variation and low-contrast boundaries further weaken their visual cues. Fine positional and boundary information is also prone to attenuation during successive downsampling in the YOLO11 backbone. To address these issues, this study proposes MPS-YOLO, an enhanced YOLO11n model for sugarcane stem node detection. A public sugarcane stem node dataset containing field and indoor images was used for model training and benchmark evaluation. First, a Node Morphology Detail Enhancement (NMDE) module is introduced into the shallow backbone to jointly extract horizontal, vertical, local-texture, and Sobel-gradient features, with channel-wise gating used to recalibrate fused representation. Second, the DySample operator is incorporated into the high-resolution feature fusion path to reconstruct fine spatial features through learned sampling offsets. Third, a high-resolution prediction branch is added to the neck to retain positional and boundary information for densely distributed sugarcane stem nodes. Under a fixed data split and unified training settings, the results averaged over five runs with different random seeds showed that MPS-YOLO improved precision, recall, [email protected], and [email protected]:0.95 over YOLO11n by 2.00, 2.86, 2.08, and 4.50 percentage points, respectively. The proposed model has approximately 2.7 M parameters and requires 11.8 GFLOPs. These results demonstrate improved detection and localization on the public benchmark and may support visual localization in mechanized seed-cane cutting.

Frontiers in Plant ScienceVol. 17
Guangxi Normal University (CN), Hunan Vocational Institute of Technology (CN), Xiangtan Vocational College of Science and Technology (CN), Nanning University (CN), Guangxi Science & Technology Normal University (CN)
National Natural Science Foundation of China, Guangxi Key Research and Development Program
Zero hunger
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon