Vine nutrition
Abstract
A smartphone app was developed for the diagnosis of nutritional disorders in grapevines using visual leaf symptoms. The app uses artificial intelligence (AI) to assess images captured by the user using a standard camera on a smartphone. To develop the symptom database, glasshouse Chardonnay and Shiraz vines were grown hydroponically in various nutrient solutions and RGB images were captured of leaves as symptoms developed and progressed. The image database includes symptoms for nitrogen, potassium, magnesium, calcium and iron deficiency. Machine learning was used to process the images and a model with high accuracy and rapid processing times was incorporated into the app. However, outdoor image acquisition of leaves on field vines resulted in lower accuracy and further development is therefore necessary. In a separate study, hyperspectral imaging accompanied by machine learning also proved effective in identifying leaf age-based differences and individual nutrient disorders. Lastly, based on gradients in nutrients along the petiole, tissue sampling protocols for nutrient assessment were refined.
Summary
Optimum grapevine nutrition is required for long-term plant resilience against biotic and abiotic stresses, and for achieving the desired yield and berry composition. The most accurate method to determine vine nutrient status is through a chemical analysis of sampled tissues. Obvious visual leaf symptoms can also be useful in diagnosing nutrient disorders but reference images provided in field manuals are often not clear, not cultivar specific and don’t show the progression of symptoms. Better methods are required to use these visual leaf symptoms to optimum benefit. This project had 3 objectives. The first objective was to develop a smartphone diagnostic app that provides information to growers on nutritional disorders in grapevines using visual symptoms. The second objective was to assess hyperspectral imaging as a method for diagnosing early-stage nutritional disorders. The third objective was to assess various tissue sampling protocols for vine nutrient assessment.
Diagnostic App
The advancement of image processing and machine learning has made it feasible to develop rapid tools to assess grapevine nutritional disorders using visual symptoms. We therefore set out to develop a smartphone app to capture and analyse images of vine leaves displaying symptoms of particular nutritional disorders. This tool uses underlying customised machine learning and computer vision techniques to assist identification of nutrient deficiencies. The app also contains an image library of symptomatic leaves for various nutrient disorders to assist growers in making in-field assessments. Finally, the app contains links to information on how to remedy the specific nutritional disorder. Limited field testing has indicated that non-uniform light and complex backgrounds decreases the app’s accuracy and therefore further development is required. The app is useful only after visual symptoms appear and is not intended as a replacement for plant tissue tests.
Hyperspectral Imaging Study
Hyperspectral (HS) imaging is a potential method to detect early indicators of plant stress, invisible to the human eye. In this study, HS imaging was successfully employed in the 380 nm to 1000 nm wavelength range to investigate the efficacy of detecting age, healthiness and individual nutrient deficiency of grapevine leaves collected from vineyards located in central west NSW, Australia. Several features were employed across the Ultraviolet (UV), Visible (VIS) and Near Infrared (NIR) regions including mean brightness, mean 1st derivative brightness, variation index, mean spectral ratio, normalised difference vegetation index (NDVI) and standard deviation (SD). Experimental results demonstrate that these features could be utilised with a high degree of effectiveness to compare age, identify unhealthy samples and not only to distinguish between control and nutrient deficient leave but also to identify specific nutrient deficiencies. Therefore, our work corroborated that HS imaging has excellent potential as a non-destructive as well as a non-contact method to detect age, healthiness and specific nutrient deficiencies of grapevine leaves. With further development, this technique could allow frequent assessment of nutrient status across a block and allow the development of nutritional maps at several time points across a growing season.
Tissue Sampling Study
Leaf tissue nutrient concentration is useful for determining grapevine nutritional status and managing vineyard nutrition. Current Australian guidelines are based on the analysis of petioles at flowering or leaf blades at veraison, sampled adjacent to the basal inflorescence and bunch. Our data indicate that nutrient concentrations were non-uniform along the length of the petiole. The concentrations of some nutrients within the leaf blade were correlated with those of the petiole but this was dependent on the cultivar and the time of sampling. The nutrient concentrations of the bunchstem were more closely correlated to the petiole than the blade. Cultivar differences were also apparent, but this depended on the individual nutrient and the sampling time. Therefore, tissue nutrient analysis will be meaningful only if sampling is consistent across these variables. These results suggest that the entire petiole should be sampled and that vineyard specific historical databases with consistent sampling procedures will be most meaningful for nutrition assessments.
This work was carried out in a close collaboration between Regional NSW, the Department of Computing and Mathematics at Charles Sturt University, and the National Wine & Grape Industry Centre with strong support from the NSW Wine Industry. Further support was provided by the GATE during the commercialisation phase of the app.