Advanced Thermal Imaging and AI: A Breakthrough in Honey Purity Detection

Revolutionizing Honey Quality Control Through Thermal Analysis

The global honey industry faces significant challenges with adulteration, where pure honey is mixed with cheaper syrups and sweeteners. Traditional detection methods often require complex laboratory equipment and time-consuming procedures. However, recent research demonstrates how combining thermal imaging with advanced artificial intelligence creates a rapid, accurate solution for identifying adulterated honey products.

Research Methodology and Sample Preparation

The groundbreaking study utilized five distinct honey varieties sourced from different regions of Morocco between April and June 2023. The selection included two thyme varieties (Thyme-A and Thyme-B), two euphorbia varieties (Euphorbia-A and Euphorbia-B), and one thistle variety. This diverse sampling strategy allowed researchers to evaluate how geographical variations within the same honey type might affect detection accuracy across different adulteration levels.

Before thermal imaging, researchers conducted comprehensive physicochemical analysis of pure honey samples, measuring key quality indicators including Hydroxymethylfurfural (HMF) content, diastase activity, Brix percentage, and refractive index. All samples fell well within international quality standards, with HMF levels ranging from 11.94 mg/kg to 40.04 mg/kg—below both European Union and Codex Alimentarius regulatory thresholds., as our earlier report

Experimental Design and Adulteration Process

The experimental design involved preparing 84 honey samples, divided into training (56 samples) and testing (28 samples) sets. Researchers used glucose syrup with a dextrose equivalent between 40 and 60 as the adulterant, adding it to honey samples at concentrations of 1%, 3%, 5%, 10%, 20%, and 30% of total sample weight. Each sample weighed precisely 5 grams, with measurements taken using analytical scales for accuracy., according to recent research

To ensure uniform distribution, glucose-honey mixtures underwent thorough agitation until completely homogeneous. The prepared samples were then incubated for 15 minutes at 60°C before thermal imaging, creating standardized conditions for comparison across all test scenarios., according to market developments

Thermal Imaging Acquisition and Processing

The research employed a FLIR ONE PRO thermal camera operating at 8.7 Hz within a spectral range of 8 to 14 μm. This portable device, capable of attaching to smartphones via USB-C connection, featured 160×120 pixel thermal resolution and temperature detection ranging from 20°C to 400°C with 3% accuracy. Researchers intentionally used multiple smartphone models (Samsung S21 FE 5G and LG Velvet 5G) to enhance the model’s real-world applicability across different devices.

Fifteen-minute video recordings captured the cooling process of each heated honey sample. Using FFMPEG software, frames were systematically extracted at 30-second intervals from both training and testing videos. This approach generated comprehensive datasets while excluding frames with imperfections like blurring or noise that could compromise data quality.

Image Processing and Feature Extraction

The image processing pipeline began with Region of Interest (ROI) detection, focusing on identifying essential areas within each thermal image. This critical step employed advanced techniques including edge detection and image segmentation to isolate key regions while eliminating irrelevant background noise.

The process initiated with specialized filtering to enhance original image quality, followed by grayscale conversion during pre-processing. Enhanced image boundaries improved object visibility, while edge detection algorithms identified significant variations in thermal intensity, effectively highlighting outlines and object peripheries. The resulting contour maps or masks were superimposed onto original images, preserving essential information while removing superfluous elements.

To maintain input consistency, all images were standardized to 224×224 pixels. This dimensional reduction helped classification algorithms focus on the most prominent features by eliminating distracting backgrounds and irrelevant details that could dominate larger images.

Advanced AI Architecture: RegNet and CBAM Integration

The research implemented a sophisticated dual-attention mechanism integrated with a convolutional neural network (CNN) architecture specifically designed for thermal image analysis. This innovative approach combines the Regulated Network (RegNet) framework with the Convolutional Block Attention Module (CBAM) to create a highly efficient detection system.

RegNet Architecture: Unlike traditional CNNs that increase complexity through additional layers or expanded widths, RegNet employs a systematic design that scales depth, width, and complexity in a controlled, predictable manner. This structured approach, organized into stages, captures progressively abstract features while maintaining computational performance and managing overfitting.

The network parameterization uses four key parameters: initial width (number of filters in the first layer), width increment factor (controlling network growth), width multiplier (influencing later layer filters), and depth (number of network layers). RegNet’s bottleneck blocks reduce operations and parameters while preserving representational power, with residual connections mitigating vanishing gradient problems and enhancing training stability.

CBAM Enhancement: The Convolutional Block Attention Module operates through two sequential submodules—channel attention and spatial attention. The channel attention module refines feature maps by emphasizing informative channels while suppressing less relevant ones. This is achieved through global average pooling and global max pooling across spatial dimensions, followed by fully connected layers and sigmoid activation.

The spatial attention module then identifies “where” to concentrate within each image, creating a comprehensive attention mechanism that significantly improves feature representation for adulteration detection.

Implications for Food Safety and Quality Control

This research represents a significant advancement in food authentication technology, offering several practical benefits:

• Rapid Detection: The method provides results in minutes rather than hours or days required by traditional laboratory methods
• Non-Destructive Testing: Thermal imaging preserves sample integrity, allowing further analysis if needed
• Cost Effectiveness: Portable thermal cameras and smartphone compatibility reduce equipment costs
• Scalability: The approach can be adapted for various food products beyond honey
• Real-World Application: Testing across multiple devices ensures practical implementation in diverse settings

The combination of thermal imaging with advanced AI architectures like RegNet and CBAM creates a powerful tool for combating food fraud. As adulteration techniques become more sophisticated, such technological innovations provide crucial defenses for protecting consumers and ensuring product quality throughout the food supply chain.

This methodology not only addresses current honey adulteration challenges but also establishes a framework that could be adapted for authenticity verification across numerous food products, potentially revolutionizing quality control processes throughout the food industry.

This article aggregates information from publicly available sources. All trademarks and copyrights belong to their respective owners.

Note: Featured image is for illustrative purposes only and does not represent any specific product, service, or entity mentioned in this article.