VddNet: Vine Disease Detection Network Based on Multispectral Images and Depth Map

TL;DR

This paper introduces VddNet, a novel deep learning architecture for early vine disease detection using multispectral images and depth maps from UAVs, outperforming existing models in accuracy.

Contribution

The paper presents an automatic orthophotos registration method and a new deep learning model, VddNet, specifically designed for vine disease detection from multispectral UAV imagery.

Findings

VddNet outperforms SegNet, U-Net, DeepLabv3+, and PSPNet in accuracy.

The system effectively uses multispectral data and depth maps for disease detection.

Advantages over direct UAV image methods include higher accuracy and robustness.

Abstract

Early detection of vine disease is important to avoid spread of virus or fungi. Disease propagation can lead to a huge loss of grape production and disastrous economic consequences, therefore the problem represents a challenge for the precision farming. In this paper, we present a new system for vine disease detection. The article contains two contributions: the first one is an automatic orthophotos registration method from multispectral images acquired with an unmanned aerial vehicle (UAV). The second one is a new deep learning architecture called VddNet (Vine Disease Detection Network). The proposed architecture is assessed by comparing it with the most known architectures: SegNet, U-Net, DeepLabv3+ and PSPNet. The deep learning architectures were trained on multispectral data and depth map information. The results of the proposed architecture show that the VddNet architecture achieves higher scores than the base line methods. Moreover, this study demonstrates that the proposed system has many advantages compared to methods that directly use the UAV images.