Resumen:
Nitrogen supply to plants is one of the essential preconditions for quality and balanced yields. At present, the nitrogen content is evaluated by destructive laboratory methods, which are expensive and time-consuming. This paper describes a novel methodology of non-destructive nitrogen content detection, using image data acquired with near-infrared (NIR) and visible imaging. Leaves of apple trees were imaged in situ using unmanned aerial vehicles (UAVs). NIR and visible images were used as an input to a Keras sequential model convolutional neural network. A pre-trained model VGG16 was used, with the last four layers tuned. We achieved an average accuracy of 97.9%, sensitivity of 98.6%, and specificity of 97.2% on 2,122 images of optimal and 2,176 images of low nitrogen content leaves.
