Shoppers may soon be able to carry a device the size of a mobile phone that can measure the ripeness of fruit. This is just one of the innovations arising from research into the "electronic nose" which uses advanced microchips that imitate what the human nose does.

These chips are in turn being connected to "neural networks", computer systems that can "learn" as they are used. These provide the real power behind electronic nose devices, explained Dr Evor Hines of the School of Engineering at the University of Warwick in Britain.

"The school has been involved in the development of electronic noses since the 1970s," Dr Hines said. "Ideas have progressed and we have collaborated on a number of projects." Warwick developed a sniffer for the brewing industry to monitor fermentation of beer and has developed sniffers for perfume and wine.

"One of the landmark projects in 1993-94 showed that you could use this technology to identify bacteria." Warwick joined with the nearby Hartland Hospital to build an electronic nose that could identify E coli and Staphylococcus A with between 80 per cent and 90 per cent accuracy.

Its most recent project involved the development of a sniffer that could identify the ripeness of bananas and apples. It included Warwick's Prof Julian Gardner and Dr Hines and Spanish researcher, Dr Edwardo Llobet and was part funded under the EU's ASTEQ programme for food quality monitoring.

The project included collaboration with US sniffer manufacturer, Cyrano Sciences of California. Cyrano grew out of technology developed at the California Institute of Technology and Prof Gardner is a member of its scientific advisory board.

"We thought it would be interesting to see whether we could use a nose to determine how ripe, for example, bananas were," Dr Hines said. Banana ripeness is grouped into seven categories and fruit is placed in a category after destructive testing. The challenge was doing this without destroying the banana.

The researchers were able to use an off-the-shelf electronic nose for the project, one with sensors that gave a broad response range. The ideal sensor would respond to the widest possible range of smells, but the broader the range, the more complex the computer analysis, Dr Hines explained.

Sensors can also have a narrow response range but these could only be used to identify a limited range of odours. The researchers have to balance these two extremes and develop a system that can do the job without costing too much or demanding too much computer power.

To understand how these sensors work you have to understand what a smell is. Odours are actually chemicals that come off the surface of a substance spontaneously, through evaporation or reaction with the air. These chemicals in turn cause a "signal" when they reach the olfactory cells in our noses, and which is then relayed to the brain. It learns through experience to identify individual smells on the basis of the complex signal coming from the olfactory cells.

The electronic nose imitates this, although at a much cruder level. The olfactory cells are replaced by a microchip coated with a special polymer. It also has an array of exposed tin oxide surfaces which are connected to the electronics underneath.

When the chemicals we would call a smell are taken up by a probe and moved across the microchip they alter the electrical potential and conductivity between the tin oxide surfaces which in turn produces a complex signal. Different smells have different chemical compositions and so produce different signals when passed across the microchip.

The signal is sent to a small computer for interpretation. The Warwick team used a type of processor known as a neural network which can learn as it goes along. The system isn't trying to give a yes or no response to the user, it is trying to interpret the signal and make a choice and the neural network can improve its response as it gains experience.

Initially the researchers tested bananas to see if the odours they produced clustered together into the seven ripeness categories and they found that indeed they did, Dr Hines said. "We then trained the neural network to recognise which of the seven states the fruit was in."

The researchers produced prototype sniffers for bananas and for apples. Each test takes only a few seconds, but the team achieved an accuracy of more than 92 per cent, Dr Hines said.

He sees great potential for these devices, both for the consumer and in other applications such as medicine and industry.

Warwick is already working to produce a handheld version of its fruit sniffer, one that could be taught the user's preferences for ripeness. The plan is to have a commercial prototype ready in two to three months.

They could be priced as low as £20 (€25.39), he believes, well within the reach of the consumer on a trek to the supermarket. This is a far cry from the cost of current technology. You could use a large lab based mass spectrometer to analyse smells, costing about £100,000 or a cheaper desktop PC based electronic nose at £10,000 or up.

The latest general purpose Cyrano nose costs between £2,000 and £3,000, but mass production of a consumer product would bring these costs down rapidly, Dr Hines believes.