An RGB Image Dataset of Seed Germination Prediction and Seed Vigor

Citation Author(s):
Shenyang Aerospace University
Submitted by:
chengcheng chen
Last updated:
Wed, 03/06/2024 - 04:18
0 ratings - Please login to submit your rating.


Computer vision (CV) techniques help to perform non-destructive seed viability detection (SVD) for faster, more efficient and fairer results. However, the seed vigor dataset currently suffers from insufficient number of samples, data noise, and imbalance of positive and negative samples. In order to compensate for the shortcomings of the dataset, we created a maize seed germination dataset with multi-labeled classes and sufficient sample size, which helps in modeling seed germination prediction, seed viability classification, seed viability detection, and seed germination counting.


In the zip package, two folders are included. One folder is the original images of the seed germination data with 120 folders, numbered according to serial numbers 1-120. The other folder is Seed germination data in .XML format marked by LabelImg with 120 folders, corresponding to original serial number tags 1-120.

The labeled seed germination data are classified into five categories with .XML format, which are:

1. ungerminated

2. germinating

3. germinated

4. primary root

5. secondary root