Related Experiment Video
Updated: Jul 1, 2025
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.3K
Accurate and fast implementation of soybean pod counting and localization from high-resolution image
Zhenghong Yu1, Yangxu Wang1,2, Jianxiong Ye1,3
1College of Robotics, Guangdong Polytechnic of Science and Technology, Zhuhai, China.
Frontiers in Plant Science
|March 6, 2024
Summary
A new deep learning model, PodNet, efficiently counts soybean pods for improved yield prediction. This lightweight model offers high accuracy and speed, making it ideal for real-time agricultural applications.
Area of Science:
- Agricultural Science
- Computer Vision
- Deep Learning
Background:
- Soybean pod count is a key indicator of crop yield.
- Existing models struggle with accuracy and efficiency due to pod density and distribution challenges.
Purpose of the Study:
- To develop a lightweight and efficient deep convolutional network for accurate soybean pod counting.
- To address the limitations of current models in real-time agricultural applications.
Main Methods:
- Designed PodNet, a deep convolutional network with a lightweight encoder and efficient decoder.
- Utilized a high-resolution dataset of soybean pods from field harvesting for evaluation.
Main Results:
- PodNet achieved an R² of 0.95 for soybean pod quantity prediction.
- The model has only 2.48M parameters, significantly fewer than state-of-the-art models.
- PodNet demonstrated a much higher Frames Per Second (FPS) compared to existing models.
Conclusions:
- PodNet offers a significant enhancement in efficiency with minimal accuracy compromise.
- Its lightweight architecture and high FPS make it suitable for real-time dense object counting in agriculture.
