STEMMER IMAGING

Supermarkets hold the key for many produce growers. Not only will they closely specify the size, shape and appearance of the vegetables that they will accept, they often specify a year or more in advance the amount of produce that they will want in a given week. This presents growers with significant logistical problems, as they have to plan their crops according to these future demands, yet have no control over environmental conditions such as the weather.

A vision-based harvesting system can be set-up to harvest just the part of the crop that meet the supermarket criteria, saving on subsequent sorting, but just as importantly can generate size and shape data on the crop that isn’t harvested so that the grower can combine that with short term weather forecasts to see if more of the remaining crop is likely to ripen sufficiently in the coming weeks to meet the criteria, thus reducing waste.

Mark Williamson, Director - Corporate Market Development at STEMMER IMAGING explains: “Supermarkets have attracted a lot of criticism over their policies of selling only ‘perfectly formed’ fruit and vegetables. This can lead to high levels of food wastage if a significant proportion of the harvest has to be discarded because the criteria are not met. This heaps further financial pressures on growers who are already being squeezed on profit margins, yet these are the conditions which growers have to accept.

In principle, vision is the perfect tool for this type of application considering that it is used to measure size, shape, colour etc. on a wide range of products in a factory environment. However, operating a vision system on some sort of robot vehicle in a wide variety of weather conditions is quite different to a factory. It throws up a host of challenges since the system needs to be able to operate with equal efficiency whether it is sunny, cloudy, raining, or even at night.”

Another major challenge to be overcome is that fact that fruit and vegetables are organic products which will have a number of natural variations even within those that meet the required shape and size specification. The vision system needs to be able to take that into account and STEMMER IMAGING has been making use of its CVB Manto software. CVB Manto is a versatile advanced pattern recognition tool that uses a neural technology that emerged from research in the field of Artificial Intelligence. It uses all image information from multiple image planes and can be applied to monochrome, colour and 3D images. It automatically identifies the key features that help to identify the object’s class.

Essentially the system learns to identify the patterns of interest from a set of training images using a new type of multi-scale pre-processing filter which allows the recognition of organic forms and textures. It then makes a classification choice for each inspection image it receives and calculates a confidence factor for the classification. Crucially, CVB Manto has the ability to use any number of training images, thus offering the ability to classify objects with an accuracy not possible with conventional pattern recognition tools, yet provides the flexibility to allow for the natural variations. For some applications, laser scanning 3D sensors such as the LMI Gocator are used for the measurements, with the associated monochrome 2D image allowing additional decisions to be made on surface discolorations.

Williamson concluded: “While these interesting projects make use of existing technology, there are clearly no ‘out of the box’ solutions. We have applied our extensive imaging expertise to adapt to the individual needs of different growers. Many of these projects have completed feasibility studies and are being scaled up for field testing (quite literally!)”.