Detection of Chili Foreign Objects Using Hyperspectral Imaging Combined with Chemometric and Target Detection...

Area of Science:

• Food Science and Technology
• Spectroscopy
• Computer Vision

Background:

• Chili peppers are vulnerable to contamination with visually similar foreign materials during processing.
• Manual inspection and traditional sorting methods are inadequate for detecting these contaminants, posing risks to consumer health.
• Ensuring food safety in chili products requires advanced detection technologies.

Purpose of the Study:

• To develop and validate a system for enhanced detection of visually similar foreign materials in chili peppers.
• To improve the accuracy and speed of foreign object identification and localization.
• To ensure the safety and quality of chili pepper products through advanced technology.

Main Methods:

• Hyperspectral imaging was used to capture detailed spectral information from chili samples.
• Spectral pattern recognition algorithms (Random Forest, SVM, Minimum Distance) were employed for image enhancement and pixel classification.
• Object detection algorithms (R-CNN, Faster R-CNN, YoloV5) were utilized for recognizing and localizing foreign objects in enhanced images.

Main Results:

• The Random Forest algorithm achieved up to 86% recognition accuracy for pixel samples during image enhancement.
• YoloV5 demonstrated a foreign object recognition rate exceeding 96% with a detection time of approximately 12 ms.
• The integrated system successfully detected various foreign objects, including red stones, plastics, fabrics, and paper.

Conclusions:

• The combination of hyperspectral imaging, spectral pattern recognition, and object detection offers a robust solution for detecting challenging foreign objects in chili peppers.
• This technology provides a valuable theoretical framework for real-time, batch detection systems in the food industry.
• The methodology holds potential for application in identifying foreign materials in other particulate food products.