VinEye

Accurately detecting and localizing vineyard disease detections are essential to reduce production losses. A great variety of scien- tific work focuses on remote sensing methods, while with current learning-based techniques, a continuous paradigm shift is happen- ing in this domain. Based on a thorough literature review, the need for a survey on remote assistance for vine disease detection was mo- tivated by the adoption of the recent machine learning algorithms. Thus, in this work, the research outputs from the past few years are summarised in the domain of grapevine disease detection. A re- mote sensing-based distance taxonomy was introduced for different categories of detection methods. This is relevant for differentiating among the existing solutions in this domain, the resulting methods being grouped according to the proposed taxonomy

Vine disease detection is considered to be one of the most crucial components in precision viticulture. It serves as an input for several further modules, including mapping, automatic treatment and spraying devices. In the last few years, several approaches were proposed for detecting vine disease based on indoor laboratory conditions or on large-scale satellite images, integrated with machine learning tools. However, these methods have several limitations, including laboratory-specific conditions or limited visibility into plant-related diseases. To overcome these limitations, we propose a low-altitude drone flight approach through which we generate a comprehensive dataset about various vine diseases from a large-scale dataset in Europe. The dataset contains typical diseases such as downy mildew or black rot affecting the large variety of grapes including Muscat of Hamburg, Alphonse Lavallée, Grasă de Cotnari, Rkatsiteli, Napoca, Pinot blanc, Pinot gris, Chambourcin, Fetească regală, Sauvignon blanc, Muscat Ottonel, Merlot, and Seyve-Villard 18402. The data set contains 10,000 images and more than 100,000 annotated leaves, which were verified by viticulture specialists. Grape bunches are also annotated for yield estimation. Further, we tested state-of-the-art detection methods on this dataset focusing also on viable solutions on embedded devices, including Android-based phones or Nvidia Jetson boards with GPU.

Close range unmanned aerial vehicle image acquisition.

We used Roboflow API to manually annotated the images, marking all the bacterial and fungicidal diseases on the leaves, besides the visible grape bunches (regardless of their health status). Then, we evaluated the quality of the dataset using YOLOv8 and FPN. An example of the YOLOv8 detection:

With the widening of the Agriculture 4.0 era, the use of autonomous robots in the agriculture field is becoming a priority. The key component of such an autonomous, often multi-robot system is the perception of the environment, which is based on 2D and 3D cameras. A base processing part of the 2D images is the segmentation of different zones in the images. This is the case also in the vineyards where in order to process complex plant canopies, segmenting the parts of the image containing the area of interest is a part of the pre-processing chain. In this work, we present a Feature Pyramid Network-based grape canopy segmentation method, which has great potential to create a segmentation mask, containing only the leaves and fruits of interest. We conducted our tests in different vineyards and we also obtained the above state-of-the-art segmentation results on public and custom datasets.

The problem of vine disease detection (VDD) was addressed in a number of research papers, however, a generic solution is not yet available for this task in the community. The region of interest segmentation and object detection tasks are often complementary. A similar situation is encountered in VDD applications as well, in which crop or leaf detection can be done via instance segmentation techniques as well. The focus of this work is to validate the most suitable methods from the main literature on vine leaf segmentation and disease detection on a custom dataset containing leaves both from the laboratory environment and cropped from images in the field. We tested five promising methods including the Otsu’s thresholding, Mask R-CNN, MobileNet, SegNet, and Feature Pyramid Network variants. The results of the comparison are available in Table I summarizing the accuracy and runtime of different methods.