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

VGG16-Based Vision Framework for Plant Identification and Precision Irrigation in Robotic Horticulture.

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

Citation:  2026 ASABE Annual International Meeting  2600338.(doi:10.13031/aim.202600338)
Authors:   Jannatul Ferdows Nipa, 1
Keywords:   Precision irrigation, Automatic plant identification, Convolutional neural network (CNN), VGG16, Image classification

Abstract. Automatic plant identification is an important component of precision irrigation and water management in automated robotic horticultural systems. By accurately identifying crop species, it is possible to develop irrigation practices that are appropriate for the individual crops, that suit water needs of the species and therefore optimize use of water used while achieving an optimal growth capacity of the species. This work is in greenhouse at Technical University of Munich, Freising, Germany. The experimental platform included the FarmBot open-access CNC robotic system allowing autonomous image acquisition of the system for data collection and irrigation control to determine the system automation. The robot infrastructure included an active cultivation field of about 3.0 m x 1.5 m, with cross-axis X, Y, and Z area coverage for optimal observation. Eight plants (vegetable, herb) were selected as experimental crops. USB RGB camera for high resolution images were taken utilizing the web interface of FarmBot. The collection had occurred every 24 hours, resulting to an average of around 900 images in total data collection period. The control framework used for the robotic system comprised an Arduino microcontroller for actuation and a Raspberry Pi integrating to manage system and web interface of web interfaces. For plant classification, a convolutional neural network (CNN) based on the VGG16 architecture was trained and evaluated. Performance assessment was conducted under two conditions: using the original dataset and employing data augmentation techniques. The VGG16 model trained on the unaugment dataset attained a classification accuracy of 96%, with macro-averaged precision, recall, and F1-score of 94%, 96%, and 95%, respectively. In conclusion, the findings support the effectiveness of using VGG16 CNN construct for automatic plant species recognition in robot driven horticulture. The proposed approach lays the technical framework for sustainable, species-specific irrigation intelligent systems, thus promoting sustainable agricultural practices.

(Download PDF)    (Export to EndNotes)