This dataset is designed for few-shot learning in the context of plant disease detection, specifically targeting tea anthracnose, tea brown blight, cotton fusarium wilt, and cotton powdery mildew. The dataset is a combination of self-collected data from tea gardens in Ya'an City, Sichuan Province, China, and publicly available disease image resources from the internet. It aims to support the development and evaluation of few-shot object detection models in agricultural applications.

Dataset Structure

The dataset is divided into base classes (cotton diseases) and novel classes (tea diseases). It includes:

Base Classes: Cotton fusarium wilt and cotton powdery mildew.
Novel Classes: Tea anthracnose and tea brown blight.

Data Collection

Self-Collected Data: Images of tea diseases were captured using a 2-megapixel macro camera with sufficient lighting conditions. The dataset includes real scene images with complex field backgrounds.
Public Data Integration: Open-source disease image resources were integrated to enrich the dataset and improve its diversity.

Data Preprocessing

Blurred Image Removal: Images with low Laplacian variance (<120) were removed to ensure clarity.
Redundant Image Removal: Images with high structural similarity (>0.85) were filtered out to avoid redundancy.
Data Augmentation: Photometric transformations and uniform noise addition were applied to expand the dataset.

Dataset Statistics

Disease Type	Original Count	After Cleaning	After Augmentation
Tea Anthracnose Disease	25	19	-
Tea Brown Blight Disease	28	21	-
Cotton Fusarium Wilt Disease	185	68	408
Cotton Powdery Mildew	258	64	270

Dataset Usage

This dataset is intended for few-shot object detection tasks in agricultural disease detection. It supports the training and evaluation of models that can efficiently detect diseases with limited labeled data.

Licensing Information

This dataset is released under the MIT License. You are free to use, modify, and distribute the dataset for any purpose, including commercial use, as long as you include a copy of the MIT License.

Citation

If you use this dataset in your research, please cite the following paper:

@misc {tanke_2025,
    author       = { {tanke} },
    title        = { leaf_disease_detection (Revision 33ecd0f) },
    year         = 2025,
    url          = { https://huggingface.co/datasets/ttkqwe123/leaf_disease_detection },
    doi          = { 10.57967/hf/4707 },
    publisher    = { Hugging Face }
}

Contact Information

For any questions or inquiries regarding the dataset, please contact [202205605@stu.sicau.edu.cn].

Downloads last month: 7