LADD: A Multimodal Image–Sensor Benchmark Dataset for Litchi Freshness and Formalin Adulteration Detection

The LQADD dataset is a multimodal resource designed to advance research in machine learning (ML), deep learning (DL), and computer vision (CV) for food quality assessment and adulteration detection. It focuses on litchis (Litchi chinensis), a commercially valuable tropical fruit prone to spoilage and adulteration practices.

This dataset comprises 3,000 high-resolution images and corresponding sensor measurements, captured under controlled experimental conditions across three categories:

1. Fresh Litchi (1000 samples):
- Naturally ripened fruits with optimal freshness.
- Sensor profiles reflect high moisture (75–82%), firmness (2–4 kg/cm²), moderate acidity (pH 4.5–5.0), and low volatile organic compounds.

2. Rotten Litchi (1000 samples):
- Fruits exhibiting microbial decay, browning, and texture softening.
- Sensor data characterized by reduced moisture, low firmness, elevated ethanol and CO₂ levels due to fermentation, and lowered pH.

3. Formalin-Mixed Litchi (1000 samples):
- Fruits treated with a hazardous preservative solution (90% water + 10% raw formalin) to artificially extend shelf life.
- Sensor readings show abnormally stable moisture and firmness values, with detectable formaldehyde concentrations (2–5 ppm) and suppressed natural degradation indicators.

Data Modalities
1. Image Data: 3,000 RGB images (1000 per class).
- Captured under uniform lighting and background for consistency.
- Useful for developing CNN-based classification, object detection, or hyperspectral extensions.
2. Sensor Data: Includes Moisture (%), Firmness (kg/cm²), pH, Total Soluble Solids (TSS in °Brix), Ethanol (ppm), CO₂ (ppm), and Formaldehyde (ppm).
- Provides quantitative ground truth for freshness, spoilage, and adulteration.
- Enables multimodal learning approaches that combine visual and sensor features.

Applications:
1. Food quality and safety monitoring using ML/DL models.
2. Adulteration detection (formalin and chemical preservative identification).
3. Computer vision-based fruit classification (fresh vs. rotten vs. adulterated).
4. Sensor fusion frameworks combining visual and chemical signatures.
5. Explainable AI in food safety, by linking sensor data to visual features.

Significance:
This dataset provides a unique benchmark for real-world food safety applications, bridging the gap between visual recognition and sensor-based analysis. It supports the development of robust AI-driven systems for smart agriculture, supply chain monitoring, and consumer protection against hazardous adulteration.