Harvesting vision technology for the future


Allan Anderson, Chairman of UKIVA, explains why now is the time for the agri-food sector to invest in machine vision technology.


No-one knows exactly what the ‘new normal’, post COVID-19, is going to look like. However, the virus has brought into sharp focus the continuing huge dependence on human workforces in many industry sectors and the problems arising if they are too ill to work or cannot work in close proximity to each other. This should open up many possibilities to increase levels of automation, utilising both machine vision and robot technology.


Helping the agricultural sector

The UK has traditionally relied on workers from East European countries to harvest crops such as strawberries, salads and vegetables, but this year their availability has been significantly restricted due to lockdowns and travel restrictions across Europe. The shortfall seems to have been filled through the ‘Pick for Britain’ campaign (pickforbritain.org.uk) from the ranks of UK workers on furlough although they won’t have the experience and skill set of the usual harvesters. The automation of crop harvesting using machine vision techniques has been the subject of many research projects and holds a lot of promise for the future. Some UKIVA members have been involved in successful projects in areas as diverse as broccoli harvesting and post-harvest trimming of vegetables such as swedes, leeks and sprouts. Other applications have included prototype systems for picking cucumbers, lettuce, apples and sweet peppers.

The key challenge in crop harvesting is to distinguish the crop from any surrounding material such as leaves, to identify when it is ripe and ready for picking. While this is something the human brain is very good at, the subjectivity from person to person can still lead to picking before individual items are really ready, leading to significant waste. Automating the process not only requires produce recognition, but also interaction with a robot for picking, and usually involves the use of 3D imaging in combination with deep learning since 3D imaging allows the shape and size of the crop to be determined. With significant developments in lighting configurations, 3D processing power, low cost embedded cameras and greatly simplified vision-robot interfaces, 3D systems can readily be mounted on the vehicles used for harvesting. As these are generally mobile systems, the most popular 3D technique for these applications is laser line triangulation where a line is projected onto the surface of the crop. As the line moves across the crop, a series of profiles is built up to give the topology of the crop. Classification of the 3D images produced is generally carried out using deep learning methods, which utilise artificial neural networks to imitate the way the human brain works for recognition and decision making. These work particularly well for organic materials where there are lots of natural variations. The system is provided with several training images from which the classification required is learned, and then this is applied in real time to the crop in the fields. The vision system can then send picking coordinates to the robot. If the system is linked in with GPS data, the location of crop that is not quite ready yet can be stored for automated harvesting at a later date.


Now is the time to invest

Machine vision has an important role to play right across the agri-food industry. Some applications are well-established while others such as harvesting are at the earlier stages. In the aftermath of the coronavirus pandemic the time is right for greater focus and investment in this area. In the budget announced in March even before COVID-19 took hold in the UK, it was announced that investment in R&D would increase to £22 billion per year by 2024-25. Significant sums are already allocated to agri-food programmes led by Innovate UK and BBSRC (Biotechnology and Biological Sciences Research Council). It would be good to see some of this investment help lead to more vision-driven automation systems literally deployed in the field for future support of this key sector of the economy.





About Author

Comments are closed.