Related Experiment Video
Updated: Sep 15, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.6K
Multifunctional cells based neural architecture search for plant images classification.
Lin Huang1, Xi Qin1, Tiejun Yang2
1College of Physics and Electronic Information Engineering, Guilin University of Technology, Guilin, 541006, Guangxi, China.
Scientific Reports
|July 16, 2025
Summary
We developed MFC-NAS, a neural architecture search method for plant image classification. This method achieves high accuracy (99.10%) with fewer parameters and faster inference than SOTA models.
Area of Science:
- Computer Vision
- Machine Learning
- Deep Learning
Background:
- Automated plant image classification is crucial for agriculture.
- Developing high-performance Convolutional Neural Network (CNN) models requires efficient architecture search.
- Existing methods may lack efficiency in parameter usage and inference speed.
Purpose of the Study:
- To propose MFC-NAS, a novel Neural Architecture Search (NAS) method for plant image classification.
- To design a search space utilizing multifunctional cells (MFCs) including transfer, normal, pooling, and dropout cells.
- To optimize CNN architectures for high accuracy and efficiency.
Main Methods:
- Designed an MFC-based search space incorporating transfer, normal, pooling, and dropout cells.
- Employed an MFC-oriented search strategy, leveraging pre-trained models like MobileNet V3 for transfer cells.
- Explored diverse pooling types/sizes and dropout rates within the search space.
- Stacked optimal cells to construct the final plant image classification CNN.
Main Results:
- MFC-NAS identified optimal cells after approximately 69 GPU-hours of search.
- Achieved a high accuracy of ~99.10% on plant image classification tasks.
- Demonstrated significantly reduced network parameters (1.58M, 6.9% of ResNet-50) and fast inference (12.6 ms).
Conclusions:
- MFC-NAS effectively develops high-performance CNNs for plant image classification.
- The proposed method offers a superior balance of accuracy, efficiency, and parameter reduction compared to SOTA models.
- MFC-NAS represents a significant advancement in automated deep learning model design for specialized image classification tasks.
