Use of artificial intelligence on UAVs for real time plant diseases detection

Precision agriculture principles appeared in the early 1980s as field management techniques in order to improve the inputs/yields ratio such as the insertion of nitrogen, phosphorous and potassium in high energy and intensive crops (i.e. maize, sugar beet). Nowadays, precision agriculture aims to optimize investments and yields taking into account environmental and conditions variability between different plots, influencing every aspect of agriculture such as tillage, seeding, fertilization, irrigation and pesticide spraying. In order to achieve all these goals, development of new sensors and field data gathering must be taken into account. In recent years, unmanned aerial vehicles (UAVs) have been used in agriculture as part of photogrammetric and remote sensing tasks, but new opportunities arise also in pesticide distribution. In this case, high resolution images are required, introducing a technical complexity linked to data transfer and storage. One possibility is to store only images requiring a post processing, dropping all images proposing a standard healthy crop. The use of artificial intelligence (AI), and deep learning (DL) in particular, allows larger learning capabilities and thus higher performance and precision of real-time classification and detection. The aim of this work is to develop and test a DL model in order to detect in real-time plants diseases. The neural network has been trained with a data set of RGB images, then tested on a test set. The system adopted a convolutional neural network (CNN) as feature extractors from the input images and TensorFlow as framework, showing good results in disease detection.