Visual apple maturity-stage detection in complex orchard environments via color–texture fusion and weakly supervised foreground modulation

Introduction Visual detection of apple maturity stages in natural orchards is often hindered by occlusion by leaves and branches, illumination variation, background similarity, and subtle differences between adjacent visual maturity stages. Existing lightweight detectors remain limited in representing fine-grained color–texture cues and explicitly regulating spatial foreground responses under complex orchard backgrounds. Methods In this study, a YOLO11n-based model, termed DS-MOFPS-YOLO11n, was developed for visual apple maturity-stage detection under complex orchard conditions. The model employs a Color–Texture Dual Stem (CTDS) to separately extract and adaptively fuse color and texture features, thereby improving the representation of shallow visual maturity-discriminative cues. A Maturity-Oriented False Positive Suppression (MOFPS) module is introduced at the P 3 and P 4 feature levels to generate spatial foreground-response maps for feature modulation. In addition, a bounding-box-guided foreground supervision loss is designed to construct weak foreground supervision signals from existing bounding-box annotations and constrain the spatial responses of MOFPS. Results On the test set of a public apple maturity dataset, the proposed model achieved a precision of 91.6%, a recall of 89.9%, an mAP50 of 95.3%, and an mAP50–95 of 91.5%. Paired analysis across five random seeds further showed significant improvements in both mAP metrics. Ablation and foreground-response alignment analyses further examined the contributions of the proposed components and showed that foreground supervision made the MOFPS responses more concentrated on annotated apple regions. Discussion Overall, DS-MOFPS-YOLO11n improves visual apple maturity-stage detection while maintaining a compact parameter count and moderate computational overhead, although broader validation across datasets and practical edge-device evaluation is still required.

Authors

Institutions

Publication Details

Journal
Frontiers in Plant Science
Published
2026-11-16
DOI
https://doi.org/10.3389/fpls.2026.1976548
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Visual apple maturity-stage detection in complex orchard environments via color–texture fusion and weakly supervised foreground modulation

Xia Wang, Can Zhao, Yaqin Wu, Mengfan ZHANG et al.
2 links
Frontiers in Plant Science
Smart Agriculture and AI
article

Visual apple maturity-stage detection in complex orchard environments via color–texture fusion and weakly supervised foreground modulation

Xia Wang, Can Zhao, Yaqin Wu, Mengfan ZHANG, Yimin Xia, Lijun Cheng, Jia Lv
article en

Abstract

Introduction Visual detection of apple maturity stages in natural orchards is often hindered by occlusion by leaves and branches, illumination variation, background similarity, and subtle differences between adjacent visual maturity stages. Existing lightweight detectors remain limited in representing fine-grained color–texture cues and explicitly regulating spatial foreground responses under complex orchard backgrounds. Methods In this study, a YOLO11n-based model, termed DS-MOFPS-YOLO11n, was developed for visual apple maturity-stage detection under complex orchard conditions. The model employs a Color–Texture Dual Stem (CTDS) to separately extract and adaptively fuse color and texture features, thereby improving the representation of shallow visual maturity-discriminative cues. A Maturity-Oriented False Positive Suppression (MOFPS) module is introduced at the P 3 and P 4 feature levels to generate spatial foreground-response maps for feature modulation. In addition, a bounding-box-guided foreground supervision loss is designed to construct weak foreground supervision signals from existing bounding-box annotations and constrain the spatial responses of MOFPS. Results On the test set of a public apple maturity dataset, the proposed model achieved a precision of 91.6%, a recall of 89.9%, an mAP50 of 95.3%, and an mAP50–95 of 91.5%. Paired analysis across five random seeds further showed significant improvements in both mAP metrics. Ablation and foreground-response alignment analyses further examined the contributions of the proposed components and showed that foreground supervision made the MOFPS responses more concentrated on annotated apple regions. Discussion Overall, DS-MOFPS-YOLO11n improves visual apple maturity-stage detection while maintaining a compact parameter count and moderate computational overhead, although broader validation across datasets and practical edge-device evaluation is still required.

Frontiers in Plant ScienceVol. 17
Shanxi Agricultural University (CN)
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