Revolutionizing Date Classification with Industrial-Grade AI

Moving beyond the human ceiling with Ain Al-Tmoor a dual-stream deep learning system achieving 86.32% accuracy across 30 cultivar varieties at industrial scale

Industry Challenge

Scaling Quality in a Billion-Dollar Industry

Immense Volume

Processing over 60,000 truckloads per season requires speed that manual labor cannot sustain

The Human Ceiling

Human accuracy fluctuates between 70-85% and degrades further due to fatigue during long shifts

Visual Ambiguity

Varieties like Fard, Fard-Liwa, and Anwan are nearly identical, leading to batch contamination and export failures if misclassified

Intelligence That Sees the Unseen

Dual-Stream Architecture
A novel dual stream system combining DINOv3-Large Vision Transformers with domain-adaptive texture encoding
Beyond Pixels
Our model captures spatial structures and surface cues (color histograms and texture patterns) to distinguish between confusable clusters
Video-Based Inference
Real-time frame extraction at 10 FPS with majority voting aggregation for high-confidence results

How It Works

On-Site Video Capture

At the farm or receiving center, the user records a high-definition video of the date pallet. This multi-angle, 3D perspective overcomes the limitations of static 2D imaging

Secure Digital Upload

The user uploads the video to the Ain Al-Tmoor web platform and requests a classification, triggering the industrial inference pipeline on Microsoft Azure

Advanced Frame-by-Frame Processing

The model extracts frames at 10 fps, feeding them into a dual stream architecture that combines Vision Transformer spatial analysis with surface texture and color cues. A majority voting system then aggregates these results to determine the final classification

Automated Classification Output

The system identifies the date variety from 30 supported classes, providing a confidence score for reliability. Any mismatch between the prediction and the farmer's declaration is automatically flagged for two-tier human verification before routing

Model Results

Model Results & Performance

Ain Al-Tmoor sets a new state-of-the-art benchmark for fine-grained agricultural classification, specifically designed to handle the visual complexities of 30 distinct date cultivars

%
Accuracy
%
Precision for Key Varieties

Classify Your First Batch

Scroll to Top