Vision systems, sensing and sensor networks to manage risks and increase productivity in vegetable production systems (VG15024)
What was it all about?
This collaboration, which ran from 2016 to 2018, brought together expertise in engineering, robotics, machine learning, plant pathology and agronomy to investigate application of vision systems, hyperspectral imaging and wireless sensor networks in horticulture. The research involved experts from the Queensland Department of Agriculture and Fisheries (DAF), the Queensland University of Technology (QUT) and the CSIRO.
The project’s aim was to develop novel sensor technologies and algorithms that were modular and so could be used on multiple systems. The research yielded promising results in the areas of rapid yield assessment and earlier problem detection.
The project addressed several priorities that had been identified with growers and industry during the needs analysis phase of the Hort Innovation Vegetable Fund project Evaluation of automation and robotics innovations: developing next generation vegetable production systems (VG13113). These were...
Vision systems that can rapidly and accurately assess crop yields of fruiting vegetable crops
The team continued earlier work that used capsicum as the target crop for machine learning, aiming to improve on earlier results. The team captured data from capsicum crops grown in a greenhouse, as well as the more challenging field environment, where leaves can obscure fruit.
The algorithms were able to identify fruit to a high level of accuracy – 80 per cent in greenhouse trials and almost as well in the field. Performance of the system was similar for single and double row planted capsicum and gave a slightly better result when mounted on a robotic platform. The system was successful in distinguishing the colour of fruit, however it was confused by breaking colour fruit which it couldn’t distinguish from red capsicum.
The results are encouraging, suggesting potential for automated crop forecasting and ultimately, selective robotic harvesting. The results achieved by the team will be conveyed to vision system researchers to advance the technology globally. The next step is to further develop the system by field-testing with potential end users.
Imaging to detect crop problems before they are visible to growers
This work was high risk but with potentially large future pay-off for industry. Tomato spotted wilt virus (TSWV) in capsicum was selected as the case study and CSIRO and DAF researchers completed five glasshouse pot trials over a two year period.
In each trial, a proportion of plants was inoculated with the virus. In the first three trials, hyperspectral cameras scanned leaves cut from plants, which were then assessed visually for symptoms to ground-truth imaging data. In this first round of trials, CSIRO’s algorithms using leaf reflectance were able to discriminate between leaves with symptoms and healthy leaves at greater than 85 per cent accuracy.
In the last two trials, whole capsicum plants in pots were scanned and visually assessed for symptoms every two days to track symptoms as they developed over time. CSIRO achieved greater than 90 per cent accuracy in distinguishing between plants with symptoms and healthy plants.
Information from the images was better than the human eye, with the system able to correctly predict plant infection at least five days before an experienced plant pathologist observed symptoms in plants.
Results indicate that hyperspectral imaging combined with machine learning shows great promise for future automated crop health monitoring. Next steps are to investigate if algorithms are robust enough to detect TSWV in capsicum in field situations and test if they can be adapted to other crops and diseases.
Outputs from the research proved to be well targeted, demonstrating that having potential end users involved in research from the beginning is beneficial. The team communicated with growers and industry through multiple channels including annual industry forums in Bowen, Bundaberg and Gatton, regular grower levy partner webinars as well as start and end of project evaluation interviews. Local grower associations and DAF agronomists also distributed six two-page project updates (QLD Veg Automation News) through their networks to keep industry informed about the project.
This project was a strategic levy investment in the Hort Innovation Vegetable Fund