Sugarcane Node and Bud Detection Dataset
Description
The dataset for sugarcane node and bud detection features widely grown varieties, including Israr Shaheed, CPF-253, and HSF-242 from KPK, Pakistan. The purpose of this dataset collection and labeling is to train a machine learning model for accurately detecting and cutting the sugarcane buds. The dataset contains high-resolution images of sugarcane nodes and buds captured in an indoor environment with a diverse background. The dataset contains 2000 images at 1920 × 1080 resolution, captured with an A4Tech PK-940HA FULL HD 1080p autofocus webcam. Each image is manually annotated using the bounding box method based on the two labels "bud" and "node" (0: bud, 1: node). All the annotations are created using the Python-based labelling tool “Label Studio” and exported in the format of You Only Look Once (YOLO)-based deep learning models. The purpose of collecting and sharing this dataset is to promote research on automated sugarcane bud detection and planting. The dataset contains two distinct classes, buds and nodes, encoded as 0 for buds and 1 for nodes. The node class corresponds to the ring-like joint found at regular intervals alongside the talk, whereas the bud class refers to the specific growth points nested within each node along the stalk. Manual annotation of the dataset was performed using Label Studio, where bounding boxes were systematically drawn around sugarcane nodes and buds. The resulting annotations were exported in YOLO format to ensure direct compatibility with the object detection architectures, including YOLOv5, YOLOv7, and YOLOv8. A Python-based tool was used to validate the correctness and alignment of bounding boxes against the final annotated images. Figure 2 represents the output bounding boxes developed by the Python script Dataset Diversity To maximize the dataset generalization and mitigate model overfitting, multiple strategic layers of environmental and biological diversity were introduced during the data collection phase. Background Variations To minimize background bias and enhance feature extraction, the dataset includes sugarcane stalks against diverse background matrices such as wooden, rough plate, and marble surfaces. Phenotypic Diversity Sugarcanes exhibit natural variations in stem thickness, color, outer skin texture, and internode stalk length. Morphological Variations The dataset captures natural variations in bud size, shapes, orientations, and maturity levels across diverse stalk conditions, including clean surfaces, rough exterior fibers, and peeled segments. Ethics Statements The dataset contains no animal or human data, and no data was taken from social media platforms. Dataset Access and Consent Statement Researchers intending to use this dataset for research or educational purposes must obtain prior consent from the dataset authors by email. A detailed consent form is attached. Dr. Ihtisham Ali: ihtisham.ali@spcai.paf-iast.edu.pk Mr. Muhammad Adil Khan: m.adilkhan@ieee.org
Files
Steps to reproduce
The purpose of this dataset collection and labeling is to train a Machine Learning model for accurately detecting and cutting the sugarcane buds. The dataset contains high resolution images of sugarcane nodes and buds captured in the indoor environment with a diverse background. The dataset contains 2000 images of resolution 1920 * 1080 using A4Tech PK-940HA FULL HD 1080p auto focus webcam camera. Each image is manually annotated using the bounding boxes method based on the two labels bud and node (0: bud, 1: node). All the annotations are created using python-based labelling tool “Label Studio” and exported in the format of You Only Look Once (YOLO) based deep learning models. The purpose of collecting and sharing this dataset is to promote research on automated sugarcane bud detecting and planting.