engleski

Machine learning as a tool for non-destructive detection of nutrient deficiencies using RGB images

In plant nutrition, color changes on the leaves and fruits of plants indicate nutritional imbalance and deficiencies of certain nutrients. Nutrient deficiency symptoms are manifested differently on the leaf or fruit. Early detection of color changes in apple leaves and fruit would allow fruit growers to react in time and prevent further nutritional problems. Therefore, the goal of this project is to use machine learning to create a model as a tool for non-destructive detection of nutrient deficiencies in apple leaves and fruits. A custom-made RGB camera is used to collect images with nutrient deficiencies. From 5 commercial apple orchards nearby city of Zagreb (Croatia), 200 images are taken each day. Starting in September 2021 until May 2022, 4.278 images were collected. The images were processed in the annotation program (LabelImg) where the apple leaves were annotated as the following classes: apple leaf healthy, apple leaf healthy juvenile ; nitrogen, potassium, phosphorus, calcium, magnesium, iron, zinc, and manganese deficiency. Annotated images were used as "training data" for machine learning with the goal of creating a model. The most represented classes were 'apple leaf healthy' and 'apple leaf healthy juvenile' (5980 and 4885 annotations, respectively). Classes 'potassium deficiency' (12 annotations), 'magnesium deficiency' (5 annotations), 'zinc deficiency' (4 annotations) and 'phosphorus deficiency' (3 annotations) were minimally represented. The ultimate goal of the project is to apply machine learning to a rover that would be autonomously moving within apple orchards and independently detecting nutrient deficiencies on the apple in real time.

annotations, apple leaves, rover

nije evidentirano