Shao et al. (2025) ENT-YOLO: An improved lightweight YOLO for cotton organ detection in mulched drip irrigation systems in southern Xinjiang

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-12-07
• Authors: Jingcui Shao, Qingqing Zhao, Zhi Gong, Xinlei Guo, S. Geng, Zhaoyang Li, Dongwei Li
• DOI: 10.1016/j.agwat.2025.110054

Research Groups

• College of Water Hydraulic and Architectural Engineering, Tarim University, Aral, China
• Institute of Farmland Irrigation, Chinese Academy of Agricultural Sciences, Xinxiang, China
• Institute of Western Agriculture, Chinese Academy of Agricultural Sciences, Changji, China

Short Summary

This study proposes ENT-YOLO, a lightweight deep learning model based on YOLOv11n, for precise detection and spatial mapping of cotton organs (buds, flowers, bolls) in complex mulched drip irrigation systems in southern Xinjiang. The model achieves high accuracy (79.77 % mAP@0.5) with a compact size (4.2 MB), providing a foundation for intelligent water, fertilizer, and salt management.

Objective

• To develop a lightweight and accurate deep learning model (ENT-YOLO) for precise detection and spatial mapping of cotton organs (buds, flowers, bolls) in complex field environments of mulched drip irrigation systems in southern Xinjiang, aiming to reduce false and missed detections while maintaining model compactness.

Study Configuration

• Spatial Scale: Field plots in the Second Regiment of the First Division of the Xinjiang Production and Construction Corps, Xinjiang Uygur Autonomous Region, China (40°35′30″N, 79°04′40″E). Plant-level detection.
• Temporal Scale: Multi-day data acquisition from June to August 2024, covering budding to boll development stages.

Methodology and Data

• Models used:
  • Baseline: YOLOv11n
  • Proposed: ENT-YOLO, which integrates:
    • CSP-EDLAN module (replaces C3k2)
    • NWD loss (combined with CIoU loss)
    • TADDH (Task Align Dynamic Detection Head)
• Data sources:
  • Self-constructed Cotton Organ Detection Dataset (COD-DS).
  • 1456 raw RGB images (3472 × 4624 pixels or 4032 × 3024 pixels) captured using handheld smartphones (Redmi Note 12 T Pro, Xiaomi 11).
  • Images collected from 0.4–0.8 meters distance, random slightly downward viewing angles, multiple times of day.
  • Contains 24,978 annotated instances (bud-flower and boll classes).
  • Dataset split: 8:2 ratio for training and validation.

Main Results

• ENT-YOLO achieved 76.36 % precision, 73.44 % recall, and 79.77 % mAP@0.5.
• The model has a computational complexity of 8.4 GFLOPs, 2.10 million parameters, and a size of 4.2 megabytes.
• Compared to the baseline YOLOv11n, ENT-YOLO increased mAP@0.5 by 2.16 %.
• The Average Precision (AP) for the bud-flower class increased by 3.13 % (to 75.06 %), and for the boll class by 1.17 % (to 84.47 %).
• The overall model size was reduced by 19.2 % relative to YOLOv11n (from 5.2 megabytes to 4.2 megabytes).
• Spatial distribution maps of cotton organs were successfully generated using ENT-YOLO detection outputs.

Contributions

• Construction of a cotton organ dataset (COD-DS) covering key growth stages using RGB images collected under mulched drip irrigation, containing 1456 images and 24,978 annotated instances.
• Development of the ENT-YOLO detection model based on YOLOv11n, integrating CSP-EDLAN modules, NWD loss optimization, and the TADDH detection head, achieving 79.77 % mAP@0.5 with a compact model size of 4.2 megabytes on COD-DS.
• Generation of spatial distribution maps using ENT-YOLO detection outputs, providing reference information for studies on cotton organ distribution, growth stage identification, and irrigation management under mulched drip irrigation.

Funding

• Key Technological Research and Development Initiative for Priority Areas in Divisions and Cities of XPCC (No. KY2024GG09)
• Tianjin Key Laboratory of Rail Transit Navigation Positioning and Spatio-temporal Big Data Technology (Grant TKL2024B12)
• Agricultural Science and Technology Innovation Program (ASTIP) of Chinese Academy of Agricultural Sciences

Citation

@article{Shao2025ENTYOLO,
  author = {Shao, Jingcui and Zhao, Qingqing and Gong, Zhi and Guo, Xinlei and Geng, S. and Li, Zhaoyang and Li, Dongwei},
  title = {ENT-YOLO: An improved lightweight YOLO for cotton organ detection in mulched drip irrigation systems in southern Xinjiang},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.110054},
  url = {https://doi.org/10.1016/j.agwat.2025.110054}
}