Related Experiment Video
Updated: May 14, 2025
07:34
A Simple Stimulatory Device for Evoking Point-like Tactile Stimuli: A Searchlight for LFP to Spike Transitions
Published on: March 25, 2014
9.8K
LGWheatNet: A Lightweight Wheat Spike Detection Model Based on Multi-Scale Information Fusion
Zhaomei Qiu1, Fei Wang1, Tingting Li1
1College of Agricultural Equipment Engineering, Henan University of Science and Technology, Luoyang 471000, China.
Plants (Basel, Switzerland)
|April 12, 2025
Summary
This study introduces LGWheatNet, a novel deep learning model for efficient wheat spike detection and counting. It achieves high accuracy with significantly fewer resources, outperforming existing methods for precision agriculture.
Area of Science:
- Computer Vision
- Agricultural Technology
- Deep Learning
Background:
- Accurate wheat spike detection is crucial for efficient crop management and precision agriculture.
- Existing methods often struggle with resource constraints and accuracy, especially in complex field conditions.
- Need for lightweight, high-performance models for real-time agricultural monitoring.
Purpose of the Study:
- To develop an accurate and efficient wheat spike detection network for resource-constrained environments.
- To improve wheat spike counting accuracy across various growth stages.
- To introduce innovative lightweight object detection methods applicable to agricultural imaging.
Main Methods:
- Developed a novel network module, SeCUIB, integrating MobileNet and ShuffleNet.
- Proposed LGWheatNet, incorporating DWDown, SPPF, and LightDetect modules.
- Utilized a custom wheat spike dataset across multiple growth stages.
- Employed the Slicing Aided Hyper Inference strategy for large-scale image processing.
Main Results:
- LGWheatNet achieved high performance: Precision (0.956), Recall (0.921), mAP50 (0.967), mAP50-95 (0.747).
- Demonstrated superior resource efficiency with 1.7M parameters and 5.0 GFLOPs.
- Outperformed YOLO, EfficientDet, and RetinaNet in detection accuracy and resource usage.
- Significantly improved detection of small targets and edge regions with Slicing Aided Hyper Inference.
- Excelled in wheat spike counting, particularly during filling and maturity stages.
Conclusions:
- LGWheatNet offers an efficient and reliable solution for wheat spike detection and counting.
- The proposed lightweight architecture and inference strategy are suitable for resource-constrained agricultural applications.
- This work advances object detection techniques for precision agriculture, especially using drone-captured data.
