Automating Seedling Counts in Horticulture Using Computer Vision and AI

Abstract

The accelerated growth of computer vision techniques (CVT) has allowed their application in various disciplines, including horticulture, facilitating the work of producers, reducing costs, and improving quality of life. These techniques have made it possible to contribute to the automation of agro-industrial processes, avoiding excessive visual fatigue when undertaking repetitive tasks, such as monitoring and selecting seedlings grown in trays. In this study, an object detection model and a mobile application were developed that allowed seedlings to be counted from images and the calculation of the number of seedlings per tray. This system was developed under a CRISP-DM methodology to improve the capture of information, data processing, and the training of object detection models using data from six crops and four types of trays. Subsequently, an experimental test was carried out to verify the integration of both parts as a unified system, reaching an efficiency between 57% and 96% in the counting process.