The challenge of food inspection

Citation

Connolly, C. (2005), "The challenge of food inspection", Sensor Review, Vol. 25 No. 3. https://doi.org/10.1108/sr.2005.08725caa.002

The challenge of food inspection

The challenge of food inspection

Keywords: Image processing, Food industry, Inspection, Image sensors

Products within the food industry are much more challenging to measurement systems than components in manufacturing engineering. Foodstuffs present such a variety of shapes, colours and textures that it is difficult to specify the threshold of acceptability in quality control instrumentation. There are often additional requirements for delicate handling.

A couple of years ago I visited the Silsoe Research Institute, which specialises in the development of instrumentation in food and agriculture. I was impressed by the ingenuity used in the design of sandwich assembly and biscuit handling systems and in the development of a cunning gripper for the robotic handling of sticky objects for placing glace cherries on top of buns. Machine vision is often an integral part of automated food handling systems, and it is used to locate the object, and to assess its quality. It is relatively simple to carry out pass/fail inspection on an engineering component, checking the dimensions and the presence of all the necessary features. This is such a common industrial inspection task that image processing software houses have developed pattern-matching and templating algorithms that carry out the checks quickly and accurately and even handle changes of scale and orientation. Some machine vision companies have built dedicated smart cameras that can be plugged into an assembly line and set up very quickly to carry out such gauging and presence-checking tasks. But the quality inspection of slices of tomato requires a more complex decision- making process that can tolerate the natural variations of shape and colour yet recognise when those variations overstep the boundary of acceptability. Most solution designers write their own custom software for each particular task, for example, recognising a tomato slice by its colour, and then measuring its size and regularity of outline; and they set empirical pass/fail thresholds. There is only one off-the-shelf image processing tool that I am aware of that can cope with this type of problem, and that is Common Vision Blox's Manto tool, which is based on neural network technology.

Agriculture presents even more challenges to machine vision, in that inspection tasks must generally be carried out under a very wide variety of lighting conditions, from direct sunlight with its sharp shadows to dull overcast cloudy illumination. The recent development of very bright, robust, long-lifetime LED light sources may be suitable for some outdoor applications, and bright enough to outweigh the ambient variations.

For the last 10 years I have attended the annual IPOT exhibition, and I am very impressed with the rate of development of machine vision during that time. I attended in 1996 as an exhibitor of some new camera-control technology that I helped to develop at the University of Huddersfield. The software, written in C, was the result of a long programming and research effort by two PhD students, and a group of undergraduate and MSc students. In contrast, it is now possible for a single engineer to configure complex multi-camera inspection systems, thanks to the considerable progress in both hardware and software.

Ease of programming and connectivity are the areas where most progress has been made in the last couple of years in machine vision tools. The main image processing software packages such as NeuroCheck and Common Vision Blox are providing easy user interfaces and Ethernet connectivity. Standard interfacing such as FireWire, CameraLink, Gigabit Ethernet and USB2.0 are used in the design of new machine vision cameras.

All this is very good news for the end-user, who can use machine vision like any other engineering tool to get on with the job, instead of embarking on a huge long learning experience!

Christine ConnollyAssociate Editor of Sensor Review