Context & Problem
Crop disease accounts for significant agricultural yield losses globally, particularly for smallholder farmers who lack access to agronomists. LeafLens was engineered as a mobile-first diagnostic toolkit capable of detecting 41 crop diseases in offline regions, delivering immediate multilingual recommendations.
Architecture
System pipeline layout representing coordination blocks and execution pathways designed for this solution.
Implementation
- Trained a MobileNetV3-based classifier using transfer learning on 54,000 crop images, optimizing for mobile edge execution.
- Designed an offline SQLite metadata dictionary storing treatment recommendations in 11 regional languages.
- Engineered an image-compression pipeline on the mobile client, reducing network packet payloads for low-bandwidth situations.
Challenges & Trade-offs
CHALLENGE: High inference times on entry-level mobile devices.
RESOLUTION: Quantized the TensorFlow model weights from Float32 to Int8, reducing model size by 75% and decreasing local inference latency to 120ms.
Results & Impact
Quantitative metrics and verified outcomes obtained from benchmarks.
98.4%Diagnostic Accuracy
11Languages Supported
41Crop Disease Classes
Media Gallery
Dashboard screens and coordinate visualizations of the running system.
LeafLens leaf image classification process on mobile