ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Analysis of Transplanting and Early Growth Statistics for Spring Cabbage Under Greenhouse Conditions Using AI-based Image Processing

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600507.(doi:10.13031/aim.202600507)
Authors:   Md Nasim Reza, Kyu-Ho Lee, Sun-Ok Chung
Keywords:   Precision agriculture, cabbage, Edge AI, seedling counting, convolutional neural network

Abstract.

Mechanized cabbage transplanting in greenhouses improves labor efficiency and operational speed, but planting errors such as uneven placement, excessive burial, seedling tilt, and missing hills can reduce stand establishment and early growth uniformity. This study analyzed transplantation performance and early growth uniformity of spring cabbage under greenhouse conditions using a plot-scale image and a machine learning-based framework. Experiments were conducted with two transplanting systems (automatic and semi-automatic) and two seedling ages (30 and 35 days). Shortly after transplanting, handheld RGB imagery was collected to analyze planting quality by detecting missing, tilted, and buried seedlings, as well as by quantifying intra-row spacing and row alignment deviations. In the early head development stage, the RGB images were processed using a deep learning segmentation model to extract canopy features. Plot-level metrics included the number of plants, missing plants, buried plants, no transplanted plants, pixel-based equivalent canopy area, and the coefficient of variation of canopy size to characterize the transplanting uniformity. These image-derived indicators were evaluated against ground measurements of plant characteristics. The results indicated that automatic transplanting using 35-day-old seedlings resulted in fewer planting errors and more uniform canopy development. Canopy size metrics obtained from imagery showed strong agreement with the proper alignment and shape. A convolutional neural network classifier differentiated normal, tilted, buried, and missing seedlings with an accuracy of approximately 93%, while an ensemble regression model explained about 78% of the observed variability in canopy uniformity of the seedlings. In parallel, the model also detected and counted directly after transplanting, achieving an accuracy exceeding 94% under field conditions. Overall, the proposed workflow enables efficient and scalable assessment of transplanting quality and early growth uniformity with minimal manual effort. Future integration of scale references and stereo or multi-view imaging is expected to improve absolute canopy estimation and strengthen associations with yield-related traits

(Download PDF)    (Export to EndNotes)