?? AI for Blueberries: Disease Detection with Drone ??

Blueberry farming is at the heart of British Columbia's ???? agricultural success, contributing millions to the economy annually. However, there is a challenge, called: blueberry scorch virus (BIScV). This aphid-transmitted virus poses a severe threat to the health and productivity of highbush blueberries, leading to substantial economic losses for farmers.

Until now, managing this disease has required intensive manual labour and expert evaluations — methods that are neither scalable nor sustainable for large-scale farming operations.

But there’s hope on the horizon.

A groundbreaking study has introduced a state-of-the-art solution that combines artificial intelligence and UAV (Unmanned Aerial Vehicle) technology to revolutionize how we detect and manage BIScV in blueberry fields.

In the figure above???: A) The Google Earth imagery showing locations of the Cities of Pitt Meadows and Vancouver, and the rectangle indicates the general area of the blueberry fields, B) UAV imagery of Field A, and C) UAV imagery of Field B. Yellow and green dots indicate BIScV infected and healthy blueberry plants, respectively, based on visual assessment of the severity of disease symptoms indicated by digits 0–5 (e.g., a yellow point with digit 5 indicates the highest severity, a yellow point with digit 1 indicates lowest severity, and a green point with digit 0 indicates a healthy plant). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

In the figure below????: a) healthy (0 disease severity rating) and as having increasingly severe blueberry scorch disease symptoms, b) 1 disease rating, c) 2 disease rating, d) 3 disease rating, e) 4 disease rating, and f) 5 disease rating.

Understanding the Problem

BIScV has been identified as one of the most significant diseases impacting blueberry crops in Canada.

Symptoms such as leaf chlorosis, necrosis, and bush dieback are challenging to detect visually, especially over expansive farmland. Infections can reduce yields by over 85% within a few years, forcing many growers to replant entire fields.

Traditional methods of disease detection rely on visual assessments by experts—a process that is not only time-consuming but also prone to inconsistencies. With the stakes this high, the need for a faster, more accurate, and scalable solution has never been more critical.

The Game-Changer: Introducing the Scorch Mapper

Enter the Scorch Mapper—an innovative deep learning algorithm designed to map BIScV infections using high-resolution imagery captured by UAVs. This technology is a first in its field, merging the power of two advanced AI techniques: Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs).

Here’s how it works:

• High-Resolution Data Collection: UAVs equipped with advanced sensors fly over blueberry fields, capturing detailed RGB images.
• Intelligent Analysis: The Scorch Mapper processes these images, using CNNs to identify local features (like leaf texture and color) and ViTs to analyze broader contextual patterns across entire plants.
• Actionable Insights: The result is highly accurate maps that distinguish healthy plants from those infected with BIScV, enabling growers to target their management efforts precisely.

In the figure above ???: the model was developed for mapping blueberry scorch disease, utilizing two computationally efficient CNN– and CNN-ViT-based branches, where element-wise multiplication is illustrated by ⊙, element-wise addition by ⊕, and layer concatenation by ?. The CNN-ViT branch utilizes self-attention (A) and the Ghost module (B) in parallel, while the CNN branch uses identity mapping similar to the ResNet model.

What the Study Found

This revolutionary approach was tested on two blueberry fields in Pitt Meadows, British Columbia. The results were extraordinary:

?? Superior Accuracy: The Scorch Mapper achieved classification accuracies of 70.71% in Field A and 78.15% in Field B, outperforming other advanced models like ResNet, Swin Transformer, and Efficient Net.

??High F1 Scores: The model demonstrated exceptional precision in identifying infected plants, showing a 5–13% improvement over competing algorithms.

??Spatial Transferability: Unlike traditional models that need retraining for new data, the Scorch Mapper worked seamlessly across different fields, highlighting its robustness and scalability.

In the figure above???: the classification maps were developed utilizing a) the DSM extracted from point cloud overlayed on the RGB UAV image, b) 2D CNN, c) CMT, d) Efficient Former, e) Efficient Net, f) HybridSN, g) InFormer, h) ResNet, i) Swin Transformer, and j) the Scorch Mapper, respectively. Healthy and infected plants are represented by green and yellow, respectively (training data is masked). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

The figure above ??? has been developed, utilizing a) the DSM extracted from point cloud overlayed on the RGB UAV image, b) 2D CNN, c) CMT, d) Efficient Former, e) Efficient Net, f) HybridSN, g) InFormer, h) ResNet, i) Swin Transformer, and j) the Scorch Mapper, respectively. Healthy and infected plants are represented by green and yellow, respectively (training data is masked). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Practical Benefits for Farmers

The implications of this research go beyond academia—it’s a practical tool with real-world benefits for blueberry farmers:

1. Cost and Labour Savings: Automated disease detection significantly reduces the need for manual inspections, saving time and resources.
2. Enhanced Crop Health: Early detection allows growers to act quickly, removing infected plants before the disease spreads further.
3. Scalable Monitoring: The model’s ability to generalize across different fields makes it ideal for large-scale farming operations.
4. Improved Decision-Making: Accurate and timely insights enable better planning for disease management and crop protection.

Problems Not Yet Solved with the Scorch Mapper

While the Scorch Mapper shows promise, it has limitations. One major issue is its reliance on small datasets, which can lead to overfitting, where the model struggles to generalize to new fields or conditions.

Additionally, its ability to detect early-stage infections (disease severity rating of 1) is limited, as mildly infected plants often resemble healthy ones affected by other factors like pests or nutrient deficiencies.

The system also relies solely on RGB images, missing the enhanced accuracy that multispectral or hyperspectral imaging could provide.        

Challenges in Implementation

Practical deployment of the Scorch Mapper presents several challenges. High computational requirements, driven by its complex deep learning architecture, make it resource-intensive, potentially out of reach for smaller farms. Collecting labeled training data, such as geolocated infected and healthy plants, is time-consuming and requires field expertise. Moreover, environmental factors like weather can affect UAV image collection, while technical skills are needed to operate UAVs and process the data.

What's Next for Blueberry AI

The blueberry industry in British Columbia — and across the globe — is entering a new era. With technologies like the Scorch Mapper, we can protect crops, enhance yields, and secure the livelihoods of countless farmers. This is precision agriculture at its finest, harnessing the power of AI to solve one of the industry’s most pressing challenges.

Thank you for joining us in exploring this exciting development. Stay tuned for more updates on how we’re shaping the future of farming.

Reference

What's On?

If you'd like to receive the regular 'AI in Agriculture' newsletter in your inbox, simply add your email to my mailing list.

Join over 10,400 readers who enjoy weekly updates on AI advancements in agriculture!

?? Petiole Pro for Blueberry Counting

Please check the Petiole Pro blueberry QA web-tool using the link: https://blueberries-qa.petiole.pro/