AI in Agricultural Drones: Disease Detection

Agricultural drones are increasingly important in farming practices. The main reason for this is tremendous value which they generate for agriculture research and business. In our today's edition "AI in Agricultural Drones" we'll check the concept of IoT drones closer and see specific cases of using drones for disease detection (mildew on grapevine leaves, citrus greening and wheat yellow rust).

IoT Drones: Overview of Concept

In the recent research published by the team of scientists from Malaysia, there is a good overview of IoT drones and their applications (even further than regular agriculture tasks).

IoT drones combine drone technology with the Internet of Things (IoT), creating a powerful tool for real-time data collection, transmission, and analysis in agriculture.

These drones are equipped with various sensors and connected to the IoT network, enabling them to communicate with other IoT devices and systems.

?? For example, IoT drones equipped with multispectral cameras and machine learning algorithms, particularly Convolutional Neural Networks (CNN), are used for real-time monitoring of crop health. This includes detecting diseases and performing real-time analysis.

?? Automatic pesticide spraying systems are developed using IoT drones. These systems include components such as GPS, cameras, pump controllers, and flight controllers to ensure precise and efficient spraying.

??? IoT drones facilitate detailed mapping of agricultural fields. This mapping process helps in identifying the position and size of crops, detecting infected plants, and confirming their locations.

??Real-Time Downy Mildew Detection in Vineyards Using Drones

The recent paper, published by Greek researchers, presents a framework for early detection and estimation of downy mildew in viticulture using drones and embedded devices for real-time image acquisition.

Downy mildew is a plant disease caused by oomycete pathogens, characterized by yellowing leaves and fluffy, white or grayish fungal growth on the undersides of leaves, leading to reduced photosynthesis and crop yield.

It utilizes object detection Convolutional Neural Network (CNN) models for processing images and videos, with implementations on the Debina grape variety in Zitsa, Greece. The study compares different object detection models like Faster R-CNN and YOLO in terms of accuracy and speed. Researchers exploited the following framework:

• RGB cameras and drones capture images which are then processed by deep learning models to detect downy mildew spots.
• A total of 6,800 photos collected:1200 images annotated with 9800 labeled bounding boxes for training.

The practical outcome for farmers is the ability to quickly and accurately detect downy mildew in vineyards, enabling timely and targeted interventions to protect crops and optimize yield. Farmers need to have drones equipped with high-resolution cameras and the capability to upload images to the cloud for real-time processing using the deep learning model described in the framework.

??UAV Multispectral Imaging for Citrus Greening Detection

The US study investigated the use of UAV-based multispectral imagery to monitor citrus greening disease in citrus orchards.

Citrus greening (Huanglongbing) is a severe, incurable disease affecting citrus production, leading to significant yield losses.

By comparing canopy shape traits and vegetation indices (VIs) between healthy and infected trees, the research aimed to identify significant spectral and structural differences that can be used for disease detection. Research team used the following methodology:

• UAV images captured at 50m altitude with 70% overlap, calibrated for radiometric correction.
• Tree height, crown diameter, and canopy volume measured both on the ground and from UAV images.

Two distinctive outcomes were obtained at the experiment:

1. Healthy trees exhibited higher values in all VIs compared to infected trees, with RedEdge-based indices (NDRE and CI) showing the greatest sensitivity.
2. Flush ratio (percentage of young leaves) and canopy volume were effective indicators of disease presence, with healthy trees showing significantly higher values.

The use of UAV-based multispectral imagery enables farmers to accurately detect citrus greening disease early, allowing for timely intervention and management, ultimately reducing crop losses and improving yield.

??Remote sensing of wheat yellow rust

The final paper for today was published by scientists from China. It reviews the use of remote sensing technologies combined with machine learning and deep learning algorithms for monitoring crop diseases. It discusses various remote sensing platforms and sensors, and the application of these technologies in early disease detection. The review also explores self-supervised learning as a promising approach for addressing challenges in early disease detection. For example, Bohnenkamp's use of SVM for wheat yellow rust classification with 92.9% precision and Lan's use of RGB images for disease identification.

Wheat yellow rust, also known as stripe rust, is a fungal disease caused by Puccinia striiformis that affects wheat crops, leading to yellow pustules on leaves and significant yield losses.

The research team had one main challenge - need for high-quality labelled data for training models. This challenge can be avoided by reducing the need for large labelled datasets and improving model generalization.

The adoption of remote sensing and machine learning technologies enables regular farmers to precisely monitor and manage crop diseases, significantly reducing crop losses caused by wheat yellow rust.

?? What's next in AI in Agricultural Drones?

Next edition of "AI in Agricultural Drones" will be focused on precise application of agrichemicals. . Or... would you like to read something more specific?

Please, share your thoughts in the comments ??

Wishes of healthy crops and insightful drone flights,

Maryna Kuzmenko, Ph.D ????, Chief Inspiration Officer at Petiole Pro Community

#drones

Photo credit for cover image: Kontogiannis, S.; Konstantinidou, M.; Tsioukas, V.; Pikridas, C. A Cloud-Based Deep Learning Framework for Downy Mildew Detection in Viticulture Using Real-Time Image Acquisition from Embedded Devices and Drones. Information 2024, 15, 178.