AGRICULTURE g


and crop prediction. It uses advanced navigation methods and AI to perform these tasks. Clock House Farm will be the demonstration farm for the project and, focusing on the packing side, the Manufacturing Technology Centre (MTC) will use its expertise in simulation, automation and process optimisation to ensure that the facility is working as efficiently as possible, and that any automation used is in the right area and delivering value. MTC is no stranger to research in this


area, having recently made a breakthrough with a learning robot that could benefit agriculture, among a number of sectors. Te latest development allows single objects to be picked out of a random tray or bin, without the need for high-cost sensors or lengthy programming. Te new process, which has been given the codename Project Viper, uses a deep


TerraClear’s technology is fitted to a tractor or skid steer


neural network implemented using the PyTorch machine learning framework to generalise learned behaviour to new scenes. Algorithms were trained with open-source manually-labelled datasets and simulated datasets. Performance tests, using low- cost depth cameras and collaborative robots, were carried out on a wide range of objects including metal components, cosmetic containers and fruit, and the team demonstrated that 94 per cent of attempted picks were successful. MTC senior research engineer, Mark


Robson, said: ‘Building the system around a neural network architecture allows us to update the model as we gather data in operation, enabling the system’s performance to continue to improve over time. Using simulation to automate the creation of training data significantly reduced the cost and time typically required to manually produce the large quantities of data needed to train a neural network.’


‘We can manage the number of apples depending on the amount of apple blossom’


Te MTC showed that the model could


be trained on simulated data, therefore reducing the need for labour-intensive manual data collection and labelling.


Future gazing While work on automating fruit picking continues at LIAT, Pearson noted that there’s still lots of research on robot weeding. ‘We’re doing a lot of revision work for that,’ he said. Te work revolves around identifying


the crop from the weed, along with how to make the vision process stable, from one field to the next, on different days and with different crops. ‘Tese are really difficult machine vision challenges in real-world environments,’ he added. ‘What we’re having to do is really advance the state-of- the-art in vision technologies, to get to a point where we can apply them.’ As well as identifying weeds, the institute


Picking up rocks


How do you automate the removal of rocks from agricultural land? American firm TerraClear has built a solution using computer vision that maps the terrain and identifies the size and location of rocks for a more targeted approach to rock removal. Farmers spend a lot of


time removing rocks from their fields after tillage. There are mechanical solutions that churn through the soil, but these are slow and can only


be used in certain soil conditions. TerraClear’s rock


clearance solution starts with a drone surveying the field. Images from the drone are analysed using a neural network to map the location and size of rocks. The farmer then drives a tractor with a hydraulic picker mounted to it on a route provided by the drone map. Cameras are then used to identify rocks as the tractor moves through the field, so the


picker can pull them into the tractor bucket. The system uses Triton 2.3-megapixel cameras from Lucid Vision Labs. Images from the cameras are used to train the neural network, identify rocks from aerial imagery, and for real-time identification of rocks with the picking robot. According to TerraClear, cameras need to be compact, lightweight and energy-efficient for use on aerial platforms and mobile machinery.


is also working with machine vision to identify crop diseases. It has recently partnered with start-up Fotenix, which uses imaging and machine learning to design systems to monitor crop health. Te key technology for the firm is three-dimensional multispectral imaging, which allows laboratory analysis at a more economical cost, making data-driven farming more freely available. ‘We’re trying to identify [crop] diseases at a very early stage,’ explained Pearson, ‘so that they can be brought under control before they become an epidemic.’ Fotenix’s technology, which was


developed at Manchester University, uses 3D spectral imaging, sensitive from 365 to 1,050nm, to analyse plant health in the field. It can determine detailed plant health information such as nutrients and disease at a cellular scale, using different wavelengths of light to detect unique identifiers of specific plant diseases, before they’re detectable by the human eye. ‘Digital techniques are the next thing in


agriculture. Tey use real intelligence to try and transform agriculture production,’ Pearson said, adding that he expects adoption of digital technologies in agriculture to be relatively fast.


Blossoming tech Many fresh digital technologies are already making their way into the commercial


18 IMAGING AND MACHINE VISION EUROPE APRIL/MAY 2021 @imveurope | www.imveurope.com


TerraClear


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36