Data for: Generation of Annotated Image Dataset for leaf detection in Dendrobium nobile orchid

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