Data for: Generation of Annotated Image Dataset for leaf detection in Dendrobium nobile orchid
Description
The image acquisition was conducted in a controlled glasshouse at ICAR–NRCO, Pakyong, Sikkim, India during September–October, 2020. High-resolution images of Dendrobium nobile orchids (1.5–2 years old) were captured using a Nikon D5300 DSLR (EXPEED 4 processor, 24.2 MP sensor, 50 mm lens) placed on a tripod. The potted plants were manually rotated to obtain 16 images at 22.5° intervals over 360° with a black background. The orchids were grown in a medium comprising equal parts of stone/brick pieces, leaf mold, coconut husk, and semi-rotten logs, under standard horticultural management with regular nutrient inputs and integrated pest control. A total of 766 whole-plant images of Dendrobium nobile in .png format were split into training, validation, and testing sets in an 80:10:10 ratio. Of the 10,490 annotations, 8,392 were used for training, and 1,049 each for validation and testing. Manual leaf counting from multi-angle images was performed by experts. Each leaf was annotated using bounding boxes in Roboflow (https://roboflow.com/) , with fully visible leaves fully marked and partially visible ones partially annotated. A single annotator completed the task in approximately 58 person-hours. Each annotated image produced a corresponding .txt file containing the image dimensions, label name, and bounding box coordinates. This annotated image dataset will be augmented and used for training DL of YOLOV5 variant models for leaf count. Area: Agricultural Science, Deep Learning, object detection, YOLOV5
Files
Steps to reproduce
The image acquisition was conducted in a controlled glasshouse at ICAR–NRCO, Pakyong, Sikkim, India during the months of September to October 2020. Dendrobium nobile orchids aged between 1.5 to 2 years were selected for imaging under uniform horticultural conditions. Each plant was grown in a medium consisting of equal parts of stone or brick pieces, leaf mold, coconut husk, and semi-rotten logs, receiving standard nutrient management and integrated pest control practices. To maintain visual consistency and reduce background noise, a matte black non-reflective cloth of dimensions at least 2 m x 2 m was placed behind the plant, 50 cm away from the pot. High-resolution images were captured using a Nikon D5300 DSLR camera equipped with a 24.2 MP sensor and EXPEED 4 image processor, mounted on a stable tripod and fitted with a 50 mm prime lens. The camera was set at a fixed distance of 1.5 meters, aligned horizontally to the midsection of the plant, and operated in manual mode with consistent parameters: ISO between 100–400, aperture f/8 to f/11, shutter speed around 1/125 s, and daylight white balance. Each plant was manually rotated on its base in 22.5° intervals, resulting in 16 images per plant for a complete 360° view. Side, top, and close-up views were systematically captured between 9:00 AM and 11:00 AM to optimize natural lighting. Each image was in .png format. A total of 766 whole-plant images were collected and later annotated using Roboflow. Manual annotation involved drawing bounding boxes around each leaf; fully visible leaves were completely marked, and partially visible leaves were annotated as seen. A single expert annotator completed this task over 58 person-hours. Manual leaf counting was also performed from multi-angle images to ensure annotation accuracy.
Institutions
- Indian Agricultural Statistics Research Institute
- National Research Centre for Orchids
- Indian Agricultural Research Institute