Related Experiment Video
Updated: Jun 27, 2026
06:41
Imaging and Analysis for Quantifying Maize (Zea mays) Abiotic Stress Phenotypes
Published on: March 28, 2025
Lightweight Cross-Domain Few-Shot Plant Disease Recognition Through Target-Domain Statistical Calibration
Chuantao Zhao1, Ting Xu1, Zhixian Zhang1
1College of Information Science and Engineering, Shandong Agricultural University, Tai'an 271018, China.
Sensors (Basel, Switzerland)
|June 26, 2026
Summary
This study introduces a lightweight method for plant disease recognition across different datasets, improving accuracy in few-shot learning scenarios. The approach enhances model performance by adapting to new data without extensive retraining, enabling efficient mobile deployment.
Area of Science:
- Computer Vision
- Machine Learning
- Plant Pathology
Background:
- Plant disease recognition models struggle with cross-domain transfer due to data distribution gaps and limited labeled samples.
- Existing methods often fail to generalize effectively from lab conditions to real-world scenarios.
Purpose of the Study:
- To develop and evaluate a lightweight, cross-domain, few-shot plant disease recognition method.
- To address the challenge of transferring models from the PlantVillage dataset to the PlantDoc dataset.
Main Methods:
- Integration of EfficientNet-B0 for feature extraction.
- Cosine-similarity-based prototypical classification.
- Training-time target-domain Batch Normalization (BN) adaptation (TBA) using unlabeled target data for statistical calibration.
Main Results:
- Achieved cross-split mean accuracies of 42.69% (one-shot) and 54.24% (five-shot).
- Outperformed ProtoNet by 7.44% and 9.43% in one-shot and five-shot learning, respectively.
- Demonstrated that TBA significantly improves performance, with minimal gains from more complex adaptation strategies.
Conclusions:
- The proposed method offers an effective solution for cross-domain few-shot plant disease recognition.
- Training-time target-domain BN adaptation is crucial for performance gains.
- The model's core encoder shows potential for efficient mobile deployment with low inference latency on NPUs.
Related Concept Videos
02:16Light Acquisition
In order to produce glucose, plants need to capture sufficient light energy. Many modern plants have evolved leaves specialized for light acquisition. Leaves can be only millimeters in width or tens of meters wide, depending on the environment. Due to competition for sunlight, evolution has driven the evolution of increasingly larger leaves and taller plants, to avoid shading by their neighbors with contaminant elaboration of root architecture and mechanisms to transport water and nutrients.
01:25Rapid Identification of Pathogens
MALDI-TOF MS has transformed clinical microbiology by offering a rapid and reliable method for pathogen identification. The traditional approach to microbial identification typically involves time-consuming culture techniques and biochemical tests, which can delay the initiation of appropriate antimicrobial therapy. MALDI-TOF MS avoids these delays by using characteristic ribosomal protein mass patterns of microbial cells, enabling accurate species-level identification within minutes.Principle...