AI Spots Tea Leaf Diseases Before They Spread

Tea production faces constant threats from diseases that can rapidly destroy entire crops, but new artificial intelligence technology offers farmers a powerful early warning system. Researchers have developed a deep learning approach that automatically detects three major tea leaf diseases—Rust, Helopeltis, and Spider Mite—potentially saving India's crucial tea industry from significant economic losses.

The key finding demonstrates that advanced object detection models can identify specific disease patterns on tea leaves with measurable accuracy. The Faster R-CNN model achieved 25% mean average precision (mAP) in detecting disease locations, outperforming the SSD MobileNet model's 20% mAP. This means the system can correctly identify and locate diseased areas on leaves about one-quarter of the time across multiple detection thresholds, providing a foundation for practical field applications.

Researchers collected 500 high-resolution images of diseased tea leaves from gardens in Jorhat, Assam, using 4000x3000 pixel cameras in daylight conditions. The team manually labeled each disease instance with bounding boxes, then expanded the dataset to 4,500 images using augmentation techniques like rotation and random cropping to improve model robustness. They trained two object detection architectures—Faster R-CNN with ResNet50 backbone and SSD MobileNet—using transfer learning from models pre-trained on the COCO dataset, configuring them specifically for tea disease detection.

The evaluation results show clear performance differences between the two approaches. Faster R-CNN achieved 25.2% precision and 4.4% recall at intersection-over-union thresholds from 0.50 to 0.95, while SSD MobileNet reached 20.9% precision and 2.0% recall. The researchers also implemented instance segmentation using Mask R-CNN to calculate the percentage of leaf area damaged by disease, though this approach struggled with accurately masking complex leaf shapes and overlapping disease patterns.

This technology matters because tea represents one of India's major agricultural exports, with the country being the world's second-largest producer. Early disease detection can prevent the spread of destructive pathogens like Rust, which appears as orange-yellow spots and thrives in warm, humid conditions, or Helopeltis bugs that attack young shoots during peak infestation months from June to September. By identifying diseases before they spread to healthy plants, farmers can target treatments more effectively and reduce crop losses that directly impact livelihoods and national tea production.

Limitations include the models' current accuracy levels and difficulty handling complex disease patterns. The system struggles with accurately segmenting irregular leaf shapes and overlapping disease areas, and performance could improve with more diverse training data representing different climate conditions and disease severities. The researchers note that practical field implementation would require high-resolution cameras mounted on drones to capture detailed disease patterns across large tea plantations.