Review · Karaelmas Science and Engineering Journal · Vol. 8, No. 1 · 2018
Computer Vision Technology on Food Science
Summary
At a Glance
Computer vision extracts useful information about an object from images, automatically. In food plants it replaces tired human inspectors for surface defects, contamination and quality grading: fast, precise, consistent and relatively cheap.
An indispensable part of food processing for real-time quality assessment and control.
Real-time streams, robot-controlled and fully automated production lines.
Mobile hardware and software will let shoppers check quality themselves.
1 · Introduction
Why Computer Vision in Food?
- Producers must meet strict safety rules and customer demands; quality and safety control is critical.
- Many quality factors are visible: size, colour, shape and texture change as food ages or spoils, and they can be measured without destroying the product.
- The result can set the product's price or its best-before date.
- Human inspection is repetitive, tiring, costly and subjective; machines can also see beyond the visible spectrum.
Computer vision
- fast, objective, consistent
- non-destructive, low running cost
- works on the production line
Human inspector
- slow and tiring
- subjective, varies by person
- limited to visible light
2 · Computer Vision Technologies
The Computer Vision Pipeline
Each element is a pixel with a position and an intensity; analysis turns these numbers into measurements.
The object is assigned to one of a finite set of classes, automating human visual inspection and decision-making.
2 · Computer Vision Technologies · Interactive
Pipeline Lab: Grade an Apple
A synthetic image drawn in your browser, not data from the review. Lower the lighting to 60%: the shaded side, where the bruise is, drops below the fixed threshold and the bad apple passes as grade A. Lowering the threshold recovers it. Lighting is the first challenge on the next slide.
2.1 · Challenges
Three Practical Challenges
One of the factors that most seriously affects the result. Poor or uneven lighting causes unexpected and wrong decisions.
Accuracy needs enough pixels on the details that matter; high resolution costs storage and processing time.
A camera sees only a projection of one side. Fruits must be turned or viewed by several cameras to inspect the whole surface.
2 · Imaging Modalities · Interactive
Beyond the Colour Camera
Click a band. Counts are the studies in Tables 1–7 that use each technique.
3 · Applications · Tables 1–7
43 Applications at a Glance
| Product | Application | Technique | Accuracy (%) | Reference |
|---|
3 · Applications
What Each Food Group Is Judged On
Chemical, microbial and sensory quality; above all flavour, juiciness and tenderness.
Appearance and texture: skin colour, visible defects, pinkness of cooked meat.
The most perishable group: physical properties, microbial contamination, appearance.
Physical and chemical attributes; shape and texture reveal quality.
Appearance first: colour, size, shape and gloss.
Bulk density, moisture, shape and diameter drive classification.
Shape and size: variation in thickness, diameter and weight.
Visible, measurable attributes that a camera can quantify without touching the food.
4 · Future Research Directions
Where the Field Is Heading
Effective segmentation, robustness, real-time capability and standardization: good data alone is not enough.
Extra information per pixel; needs cost-effective, efficient processing to meet industry demands.
Fast, low-cost thermal, X-ray and MR imaging for internal defects, especially in fruit.
Inspection increasingly needs height and depth, not only a flat projection.
Reprogrammable automation for evaluation, handling and processing, cheaper and more flexible over time.
Large datasets and cloud computing to improve learning for quality evaluation.
5 · Conclusion
Conclusion
- With good image processing, computer vision is a proven technology for food quality and safety information.
- Colour, multispectral, hyperspectral, thermal, X-ray and MR imaging have been developed for fast, objective, consistent, non-destructive evaluation.
- Most systems use reflectance to judge surface attributes: shape, size, colour and texture.
- Computer vision helps machines locate and characterize raw materials before handling, and will remain a key research area for the food industry.
5 · Conclusion
Check Questions
- Name the five steps of a computer vision system and what each produces.
- Why can poor lighting create false defects?
- Why might a single camera miss a defect on an apple?
- When is hyperspectral imaging preferred over an ordinary colour camera?
- Which technique in the tables can see inside the product, and for what?
Show answers
- Acquisition (image), preprocessing (cleaner image), segmentation (object mask), feature extraction (size, colour, shape, texture values), classification (class or grade).
- Shadows and uneven brightness change the measured colour, so healthy dark areas look like bruises.
- It sees only one projected side of a round object; the defect may be on the hidden side.
- When the property is chemical or hidden (tenderness, contamination, adulteration, early bruises) and shows up only at particular wavelengths.
- Computed tomography (carrots) and X-ray imaging (onions): internal tissue and structure.
References
Selected References
- Sun, D.W. (2016). Computer Vision Technology for Food Quality Evaluation. Academic Press.
- Sun, D.W. (2010). Hyperspectral Imaging for Food Quality Analysis and Control. Academic Press.
- Mery, D., Pedreschi, F., Soto, A. (2013). Automated design of a computer vision system for visual food quality evaluation. Food Bioprocess Technol., 6(8): 2093–2108.
- Wu, D., Sun, D.W. (2013). Colour measurements by computer vision for food quality control: A review. Trends Food Sci. Technol., 29(1): 5–20.
- Cubero, S. et al. (2011). Advances in machine vision applications for automatic inspection and quality evaluation of fruits and vegetables. Food Bioprocess Technol., 4(4): 487–504.
See the full article for the complete reference list.