Hymilex / CropRowDetection

CropRow Detection Dataset

Here we will release the dataset for the submission (Towards agricultural autonomy: crop row detection under varying field conditions using deep learning) in the ICRA 2022.

---

Dataset Structure

The dataset is based on 500 base images. The 500 base images are augmented into 2000 sub-images. The 2000 image dataset is classified into two groups for the purpose of testing and training. Each training and testing sub-groups contains 1000 images. Each image in the dataset consists of a corresponding ground truth image.

Numeric values of the labelled coordinates are stored in .mat files. All the label coordinates in train dataset are stored in a single .mat file. The label coordinates in test dataset are stored in 10 separate .mat files.

.
│
├── Train Dataset              # 1000 images based on 250 base images. [100 Images per category x 10 Categories]
|   ├── labels.mat             # .mat file containing image label coordinates for 250 base images
│
└── Test Dataset               # 1000 images based on 250 base images
    ├── 250 Resized images     # 250 base images are resized into 512x512 resolution
    ├── 10 Data Categories     # 1000 images with 512x512 resolution
    │  ├── Horizontal Shadows  # Shadow  falls  perpendicular  to  the  direction of the crop row
    │  ├── Slope/ Curve        # Images captured while the crop row is not in a flat farmland or where crop rows are not straight lines
    │  ├── Discontinuities     # Missing  plants  in  the  crop  row  which leads to discontinuities in crop row
    │  ├── FrontShadow         # Shadow of the robot falling on the image captured by the camera
    │  ├── Dense Weed          # Weed grown densely among the crop rows
    │  ├── Large Crops         # Presence  of  one  or  many  largely  grown crops within the crop row
    │  ├── Small Crops         # Crop rows at early growth stages
    │  ├── Sunlight            # Sunlight  falling  on  the  camera  causing lens flares and similar distortions
    │  ├── Tyre Tracks         # Tyre    tracks    from    tramlines    running through the field
    │  └── Sparse Weed         # Sparsely  grown  weed  scattered  between the crop rows
    └──Labels                  # 10 .mat files containing image label coordinates for 25 base images per category

Data Augmentation

Base image is augmented into four sub images by cropping and rotating in different orientations.

Sample Data

The crop row is labelled with white lines on black background. The line width of the white line is 6 pixels. The labels could be regenerated with custom line width using the labels .mat files.

Citation

@article{de2022deep,
  title={Deep learning-based Crop Row Following for Infield Navigation of Agri-Robots},
  author={de Silva, Rajitha and Cielniak, Grzegorz and Wang, Gang and Gao, Junfeng},
  journal={arXiv preprint arXiv:2209.04278},
  year={2022}
}