Related Experiment Video
Updated: Jul 18, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
MobileNet-WDD: a lightweight deep learning image classification method for identifying defects in wheat grains.
Yunzhao Ma1,2, Wenyue Wang3, Wenfu Wu1,4
1College of Biological and Agricultural Engineering, Jilin University, Changchun, China.
Frontiers in Plant Science
|July 15, 2026
Summary
This study introduces MobileNet-WDD, a lightweight deep learning model for wheat grain quality inspection. It achieves high accuracy and efficiency, making it suitable for real-time non-destructive analysis.
Area of Science:
- Agricultural Engineering
- Computer Vision
- Machine Learning
Background:
- Automated inspection of wheat grain appearance quality faces challenges in balancing model accuracy and computational efficiency.
- Existing methods may struggle with real-time processing and resource-intensive computations.
Purpose of the Study:
- To develop a lightweight deep learning model, MobileNet-WDD, for optimizing wheat grain appearance quality inspection.
- To enhance both recognition accuracy and computational efficiency for automated wheat grain analysis.
Main Methods:
- Proposed MobileNet-WDD model based on MobileNetV4-small architecture.
- Incorporated SimAM attention mechanism for improved feature discrimination.
- Utilized Ghost convolutions and Mish activation function for network optimization.
Main Results:
- MobileNet-WDD achieved 94.2% accuracy, a 6.1% increase over the baseline.
- Reduced model parameters by 30.3% and computational cost by 27.6%.
- Increased inference speed by 1.18-fold (from 155 FPS to 183 FPS).
Conclusions:
- MobileNet-WDD offers high-precision recognition with significant lightweight and computational advantages.
- Provides an efficient and feasible solution for real-time, non-destructive wheat grain inspection.
- Demonstrates the effectiveness of integrating attention mechanisms and optimized convolutions in deep learning for agricultural applications.