CamelliaMamba-A4: Instance Segmentation and Yield Estimation of Camellia oleifera Fruits in Natural Orchards

In natural orchards, Camellia oleifera fruits can be small, occluded or touching, and similar in color to surrounding leaves. We present CamelliaMamba-A4 for fruit instance segmentation, image-level visible-fruit counting, and preliminary tree-level fresh-mass estimation. A VMamba-Tiny V2 backbone and P2–P6 feature pyramid are combined with standard-scale proposals, continuous small-object weighting and Boundary Dice supervision. The dataset contains 390 images and 4019 annotated fruit instances. Testing uses 180 correlated frames from three trees and 566 annotated fruit appearances; the tree, rather than the frame, is treated as the independent sampling unit. The reported A4 mask mAP50–95 and Recall50 are 66.273% and 86.219%, respectively, and the counting MAE and RMSE are 0.589 and 0.863 fruits per image. The evaluation groups results by tree and by object size, relative image density and occlusion category and assesses individual and joint loss removal under fixed A4 geometry across three random seeds. A preliminary yield analysis combines cross-frame unique counts with leave-one-tree-out mass calibration and recorded harvest masses from tree08 (Xianglin No. 1), tree09 (Xianglin 207) and tree10 (Xianglin 210). The harvested-fruit records contain 145 individual masses totaling 3.086 kg. Because only three independent trees are available, the tree-level intervals and yield estimates are interpreted as a pilot evaluation rather than evidence of broad orchard-level generalization.

Authors

Institutions

Publication Details

Journal
Plants
Published
2026-11-13
DOI
https://doi.org/10.3390/plants15223548
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

CamelliaMamba-A4: Instance Segmentation and Yield Estimation of Camellia oleifera Fruits in Natural Orchards

Yujia Shao, Lin Li, Wenying Zhao
3 links
Plants
Smart Agriculture and AI
article

CamelliaMamba-A4: Instance Segmentation and Yield Estimation of Camellia oleifera Fruits in Natural Orchards

Yujia Shao, Lin Li, Wenying Zhao
article en

Abstract

In natural orchards, Camellia oleifera fruits can be small, occluded or touching, and similar in color to surrounding leaves. We present CamelliaMamba-A4 for fruit instance segmentation, image-level visible-fruit counting, and preliminary tree-level fresh-mass estimation. A VMamba-Tiny V2 backbone and P2–P6 feature pyramid are combined with standard-scale proposals, continuous small-object weighting and Boundary Dice supervision. The dataset contains 390 images and 4019 annotated fruit instances. Testing uses 180 correlated frames from three trees and 566 annotated fruit appearances; the tree, rather than the frame, is treated as the independent sampling unit. The reported A4 mask mAP50–95 and Recall50 are 66.273% and 86.219%, respectively, and the counting MAE and RMSE are 0.589 and 0.863 fruits per image. The evaluation groups results by tree and by object size, relative image density and occlusion category and assesses individual and joint loss removal under fixed A4 geometry across three random seeds. A preliminary yield analysis combines cross-frame unique counts with leave-one-tree-out mass calibration and recorded harvest masses from tree08 (Xianglin No. 1), tree09 (Xianglin 207) and tree10 (Xianglin 210). The harvested-fruit records contain 145 individual masses totaling 3.086 kg. Because only three independent trees are available, the tree-level intervals and yield estimates are interpreted as a pilot evaluation rather than evidence of broad orchard-level generalization.

PlantsVol. 15(22)
Central South University of Forestry and Technology (CN), Wuxi Taihu University (CN)
Anhui Provincial Department of Education
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