README.md

---
license: cc-by-4.0
task_categories:
- image-segmentation
language:
- en
tags:
- biology
pretty_name: >-
  A dataset of the quality of soybean harvested by mechanization for
  deep-learning-based monitoring and analysis
---
# Dataset Card for Mechanized Soybean Harvest Quality Image Dataset
This dataset contains images captured during the mechanized harvesting of soybeans, aimed at facilitating the development of machine vision and deep learning models for quality analysis. It contains information of original soybean pictures in different forms, labels of whether the soybean belongs to training, validation, or testing datasets, segmentation class of soybean pictures in one dataset.

## Dataset Description
The dataset comprises 40 original images of harvested soybeans, which were further augmented to 800 images through various transformations such as scaling, rotating, flipping, filtering, and noise addition. The images were captured on October 9, 2018, at the soybean experimental field of Liangfeng Grain and Cotton Planting Professional Cooperative in Liangshan, Shandong, China. 
Each dataset contains two columns:
original_image: contains PIL of 800 JPG images of soybeans.
segmentation_image: contains PIL of 800 PNG images labeled in colors. Green means normal soybean, red means crushed soybean, yellow means impurity, and black means background.


## Dataset Sources
The images were obtained using an industrial camera during the mechanized harvesting process and subsequently annotated by experts in the field.

## Uses
The dataset is designed for:
Developing and improving online detection models for soybean quality during mechanization processes.
Analyzing soybean mechanization processes.
Training deep learning algorithms for image classification and feature extraction.

## Out-of-Scope Use
The dataset should not be employed for non-agricultural applications or outside the context of soybean quality detection during mechanization.

## Limitation
This dataset only contains original images and segmentation images for the soybean. The segmentation images are only output of the model, not the real or true classification of soybean, its background, and crashed grains. In other words, the correctness of segmentation images is not verfied by human.

## Original Dataset Structure
The dataset is structured into three main folders:
JPEGImages: Contains 800 JPG images of soybeans.
SegmentationClass: Contains PNG images with annotations.
ImageSets: Contains TXT records for data partitioning.

## Data Collection and Processing
The main goal is to combine all the files into three datasets (train, test, validation) with two columns of images. The first step is to write a csv file containing all the labels for all images.  
After that, according to the csv file, we split all the images into three folders of train, test, validation. Each folder contains two groups of files: pictureid_original.jpg, and pictureid_segmentation.jpg.
All the data processing code is uploaded in the Project1_dataset.ipynb file.
I then upload the zip file of these three folders and read those files in the load_dataset function.


## Curation Rationale
The creation of this dataset was motivated by the need for making a standardized dataset that reflects the real conditions of mechanized soybean harvesting for use in quality detection research.

## Annotation Process
Field experts annotated the dataset, manually labeling different components of the soybean images using polygonal annotations.
Bias, Risks, and Limitations
The dataset is limited to a specific soybean variety and harvesting environment, which may affect its generalizability. Future expansions are planned to include more diversity.

## Recommendations
Users should follow ethical guidelines for handling data and consider the dataset's limitations when interpreting results from their models.

## Dataset Card Authors
Man Chen, Chengqian Jin, Youliang Ni, Tengxiang Yang, Jinshan Xu contributed to the dataset preparation and curation.

## Citation
Chen, M., Jin, C., Ni, Y., Yang, T., & Xu, J. (2024). A dataset of the quality of soybean harvested by mechanization for deep-learning-based monitoring and analysis. Data in Brief, 52, 109833. https://doi.org/10.1016/j.dib.2023.109833 

## Acknowledgements
This research received partial funding from several grants from the National Natural Science Foundation of China, National Key Research and Development Program of China, and the Natural Science Foundation of Jiangsu.