MACHINE VISION BASED DECISION SUPPORT SYSTEM FOR PRECISION NITROGEN APPLICATION IN APPLE ORCHARD

Achyut Paudel

doi:10.7273/000007173

Precision management of nitrogen at the individual tree level is crucial for optimizing resource use and enhancing productivity in tree fruit crops such as apples. Traditional methods often apply uniform nitrogen rates across entire orchard blocks. This approach fails to account for the variability in nitrogen requirements among individual trees, which can be influenced by factors such as tree age, size, crop load, and local soil conditions, thus leading to over- or under-fertilization of individual trees. Consequently, there is a pressing need for rapid and accurate methods to assess the nitrogen status of individual trees efficiently across large orchards. This study developed a machine vision-based system to characterize and quantify various canopy features as indicators of tree nitrogen status. Using RGB-D cameras and image processing techniques, key canopy characteristics such as normalized canopy area and yellowness index were quantified for individual trees. The system accurately segmented the canopy of foreground trees with an F1 score of 0.78, demonstrating its ability to isolate individual trees in complex orchard environments. The normalized canopy area, representing tree growth and vigor, showed a negative correlation (r = −0.5) with experts’ recommended nitrogen application rates, indicating that trees with smaller canopies generally required more nitrogen. The yellowness index, which quantifies the ratio of yellow to green foliage during fall senescence, was calculated with an R^2 of 0.72, providing a reliable indicator of tree nitrogen status. Then, these visual canopy features were integrated into a decision support tool designed to provide specialized nitrogen management recommendations at the individual tree level. The tool also incorporated leaf nitrogen concentration, and expert opinions along with combined information from canopy features to recommend optimal nitrogen application rates. To implement this system in the real-world conditions, the decision support model running on a portable computer was then integrated with a ground vehicle equipped with localization capabilities. An autonomous navigation system using LiDAR (Light Detection and Ranging) was developed to navigate the vehicle autonomously through the orchard rows. The system utilized LiDAR point clouds to fit lines through orchard rows, correcting its heading and location for precise navigation between tree rows in the orchard. Field trials demonstrated successful navigation within the orchard with minimal manual interventions, showcasing the system’s potential for practical application. The robotic system achieved precise application of the test material (in this case water was used, for safety) within 0.51 to 1.35 m (depending on the recommended nitrogen application level) of targeted tree trunks, illustrating its capability for accurate, tree-specific nutrient delivery. This machine vision-based, automated system demonstrated the potential for rapid, non-invasive assessment of nitrogen in tree fruit orchards, which can be efficiently scaled across large apple orchards. By enabling precise, tree-specific nitrogen management, this technology paves the way for optimized fertilizer use, reduced environmental impact from nitrogen runoff, and improved fruit quality and yield in apple production. Its adaptability could also extend to assessing other key nutrients (e.g., Phosphorous) and be applied to different high-value fruit crops (such as cherries and winegrapes), enhancing overall fruit crop management practices.

MACHINE VISION BASED DECISION SUPPORT SYSTEM FOR PRECISION NITROGEN APPLICATION IN APPLE ORCHARD

Files and links (1)

Abstract

Metrics

Details