Development of a digital monitoring system for fire blight in fruit orchards

Fire blight (Erwinia amylovora) is still one of the most dangerous diseases in the fruit production of cultivated apples and pears, as it can spread epidemically and cause enormous economic damage. Regular inspections are mandatory to detect and prevent pathogen immigration and spread at an early stage of infestation. For this purpose, a high-throughput monitoring system for the detection and localization of fire blight infections in orchards is being developed in this study. The basis for digital symptom detection is the creation of a large RGB image data set with typical symptoms of fire blight infections on shoots, leaves and flowers. Since fire blight occurs sporadically under natural conditions and infested plant parts are quickly removed, the RGB images of symptoms were captured after artificial inoculation in greenhouses and experimental fields. Through image processing and machine learning, characteristic patterns of fire blight symptoms on shoots, leaves and flowers were recognized in the image data, which can be assigned to the fire blight symptoms on the infected fruit trees. The YOLO algorithm performs object detection in real time with high speed and accuracy. To train the deep learning algorithm, the shoot, leaf and flower disease symptoms on the collected photographs were manually labeled using the Computer Vision Annotation Tool (CVAT). Object detection uses bounding boxes to determine the position of the diseased objects in the image. This also enables the precise localization of disease symptoms within the orchard using a novel photogrammetric approach on georeferenced image data. Based on this data, a monitoring model will be developed that enables a high-throughput control system and continuous spatial detection and documentation of fire blight in orchards.