A repository to open rice seedling dataset.
The data descriptor was published on Remote Sensing, MDPI. (open access)
A UAV Open Dataset of Rice Paddies for Deep Learning Practice
An application of rice seedling detection using transfer learning was published on Remote Sensing, MDPI. (open access)
Rice Seedling Detection in UAV Images Using Transfer Learning and Machine Learning
- Yang, M. D.; Tseng, H. H.; Hsu, Y. C.; Yang, C. Y.; Lai, M. H.; Wu, D. H. A UAV Open Dataset of Rice Paddies for Deep Learning Practice. Remote Sens. 2021, 13, 1358. doi:10.3390/rs13071358
- Tseng, H. H.; Yang, M. D.; Saminathan, R.; Hsu, Y. C.; Yang, C. Y.; Wu, D. H. Rice Seedling Detection in UAV Images Using Transfer Learning and Machine Learning. Remote Sens. 2022, 14, 2837. doi:10.3390/rs14122837
- Rice Seedling Datasets
The data access is moved to the page Rice Seedling Dataset 水稻秧苗辨識資料集, powered by NCHU Open AI Model and Dataset Platform.
You'll be asked to sign-in for the data access, while it is free to register.
Once you have logged in the platform, direct to the webpage and press the following botton (red rectangle highlighted).
Then accept the license term to proceed the data access.
Filename File size Image size (pixels) Spatial resolution (mm/pixel) 2018-08-07_ARI80_20m_Orthomosaic.tif 465MB 19406 x 10413 5.23 2018-08-14_ARI80_20m_Orthomosaic.tif 610MB 19876 x 10687 5.11 2018-08-23_ARI80_20m_Orthomosaic.tif 557MB 18294 x 9823 5.55
These three links are the orthomosaic image of paddy No.80, TARI, which were in TWD97 / TM2 zone 121 (EPSG:3826) projection. The images were acquired in three consecutive growing stages in 2018.
Filename File size Image size (pixels) Spatial resolution (mm/pixel) 2019-03-26_ARI78_20m_Orthomosaic.tif 485MB 20192 x 10858 5.04 2019-04-02_ARI78_20m_Orthomosaic.tif 418MB 19061 x 10250 5.33 2019-08-12_ARI78_20m_Orthomosaic.tif 503MB 21998 x 11829 4.62 2019-08-20_ARI78_20m_Orthomosaic.tif 605MB 21265 x 11435 4.78 2020-03-12_ARI78_40m_Orthomosaic.tif 278MB 15933 x 8568 6.38 2020-03-17_ARI78_40m_Orthomosaic.tif 317MB 15941 x 8572 6.38 2020-03-26_ARI78_40m_Orthomosaic.tif 385MB 15962 x 8583 6.37 2020-08-12_ARI78_40m_Orthomosaic.tif 330MB 15966 x 8586 6.37 2020-08-18_ARI78_40m_Orthomosaic.tif 382MB 15977 x 8591 6.36 2020-08-25_ARI78_40m_Orthomosaic.tif 402MB 15979 x 8593 6.36
These 10 links are the orthomosaic image of paddy No.78, TARI, which were in TWD97 / TM2 zone 121 (EPSG:3826) projection. The images were acquired in 2019 and 2020.
RiceSeedlingClassification.tgz (uncompressed size 426MB)
This file includes two folders: riceseedling and arableland. Train-val and test datasets are all included.
RiceSeedlingDetection.tgz (uncompressed size 19.1MB)
This is a PASCAL VOC format object-detection dataset, which includes two folders: JPEGImaegs and Annotations.
RiceSeedlingDemo.tgz (uncompressed size 48.5MB)
This file contains 8 detection demo images and the corresponging PASCAL VOC xml format annotations.
An overview of the region of different datasets
An overview of the field no. 80 (cyan bounding area) in TARI, Taichung. Image acquired on August 7, 2018. The green bounding area represents the area for training-validation dataset, and the red bounding area represents the subsets for object detection demonstration dataset.
This dataset contains two classes:
rice seedlingarable land
An overview of image dataset
The number of images used for training, validation, and testing in the rice seedling dataset.
| Class | Training Samples | Validation Samples | Testing Samples | Total Samples |
|---|---|---|---|---|
| Rice Seedling | 22,438 | 561 | 5,048 | 28,047 |
| Arable Land | 21,265 | 532 | 4,784 | 26,581 |
| Total | 43,703 | 1,093 | 9,832 | 54,628 |
This dataset is used for object-detection model training and validation.
- PASCAL VOC xml format annotation
- 3 consecutive mission: Aug 7th, Aug 14th and Aug 23rd
- 8 subsets
- 25 samples/subset
- 600 samples total
Examples of three growth stages of the rice seedling detection dataset
The architecture of the proposed network for rice seedling classification
In the article, we proposed a simple CNN architecture which adopted the stacked convolution layer from VGG-16. The pre-trained model is provided in the model directory.
The environments for the experiments are:
- host
- Ubuntu 18.04.5 LTS Server 64bit
- NVIDIA Display Driver 450.57 (cuda 11.0)
- Docker CE 20.10.2
- nvidia-docker2
- container (image: nvcr.io/nvidia/tensorflow:20.03-tf2-py3 (from Nvidia GPU Cloud))
- tensorflow 2.2.0
- python 3.6.9
- matplotlib 3.3.0
- scikit-image 0.17.2
- scikit-learn 0.23.1
- python3-opencv 3.2.0 (apt install)
To test the provided model, simply call the tf.keras.models.load_model() and you're ready.
This dataset is used for patch-based object-detection scenario.
- 8 subsets
- 8m x 8m region
- 1527 x 1527 pixels
- PASCAL VOC xml format annotation
An overview of 8 detection demo images
| Subset No. | Raw Image | Prediction Image | Ground Truth Image |
|---|---|---|---|
| 1 |  |
 |
 |
| 2 |  |
 |
 |
| 3 |  |
 |
 |
| 4 |  |
 |
 |
| 5 |  |
 |
 |
| 6 |  |
 |
 |
| 7 |  |
 |
 |
| 8 |  |
 |
 |
Comparison of Detection Result and Ground Truth
| Subset No. | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
|---|---|---|---|---|---|---|---|---|
| Prediction | 735 | 1006 | 1037 | 809 | 1004 | 1050 | 1017 | 1032 |
| Ground truth | 898 | 1000 | 1019 | 964 | 971 | 1002 | 1033 | 1005 |
| Error (%) | 18.15 | 0.60 | 1.77 | 16.08 | 3.40 | 4.79 | 1.55 | 2.69 |