NIAB

Contact: Michael Gifford <Michael.Gifford@niab.com>

Proposal for 2022

Simon Smart <Simon.Smart@niab.com>

Analysis of satellite imagery has become a crucial part of the business of agriculture. One problem that needs solving at scale is a method to automatically and accurately delineate the areas of crops, to enable the imagery to be clipped for analysis.

Current offerings typically generate field boundaries which are not the same as the planted area are a crop. They are also usually based on old imagery. This can cause big discrepancies when used to forecast production, especially in high value crops such as potatoes and vegetables.

NIAB is seeking a service, ideally delivered via API, which can accept a single geographic point and time period, and return the crop boundary in a suitable format (e.g. GeoJSON). The image analysis system could make use of temporal changes in the imagery, which provides valuable added information. Classical image analysis should be able to solve this problem although ML classifiers would also be of interest.

Since last year, we have found an alternative solution for acquiring the data we need which combines SAR and optical data. The limiting step that remains is to generate cropped area boundaries from a single point. There are commercial services available, but they tend to limited in geographic scope, and as Charles said they are the entire field rather than the cropped area. Some that we have tested just returned openstreetmap data. It is an active area of research - see e.g. https://gtr.ukri.org/projects?ref=106000 https://www.researchgate.net/publication/343802377_A_DEEP_LEARNING_ARCHITECTURE_FOR_BATCH-MODE_FULLY_AUTOMATED_FIELD_BOUNDARY_DETECTION although this is not our area of expertise.

The account would be to https://www.sentinel-hub.com/ which provides an API to easily acquire sentinel data for specific areas of interest.

Feedback:

Thanks for the link to the Hummingbird project.

I’m reluctant to set tasks for undergraduates where they are directly competing with well-funded professional researchers. Do you think we might be able to find an alternative way of framing the problem that did not focus so much on the core algorithm, but a systems perspective that incorporates a variety of technical elements in a novel way?

For example, the joint impact of Covid and Brexit seem likely to result in major disruption of the field-to-table infrastructure that supports arable farming in the East of England. Is there potential to use crop imaging for crowd-sourced deployment of human resources in harvesting, packing and distribution?

Confirmed 2021 project: Virtual Agronomist

Background:

Using pesticides in an effective and responsible way is a critical step in both reducing the quantities used and ensuring their impact on the environment and the health of farmers and consumers is minimised.

NIAB holds one of the most comprehensive pesticide and plant protection products databases in the UK. This database is used by farmers and agronomists to search for products to use to protect their crops from pests and diseases. There are many complex restrictions around the use of pesticides including: timing of application and growth stage of the crops, pest thresholds, maximum dose rates and number of applications, to name a few.

NIAB would like to leverage this database to help advisors verify that their pesticide recommendations are valid according to label restrictions by building an ‘Agronomy engine’. The agronomy engine will accept an input consisting of a list of pesticide applications and use the information in the database to validate the input. Additionally, the engine should be able to account for approved tank mixes (combinations of pesticides in the same application) as recommended by the manufacturer.

We would like the agronomy engine to report on the geographic locations of the queries (at county level or equivalent) and the nature of the queries in a consolidated way.

Feedback:

This looks like a nice application. I think I would include a pointer to the the technical approach that might be involved - it looks as though classical expert systems methods might be an appropriate way to encode regulatory constraints, tank mixes and so on. Representing all the relevant knowledge in a way that can be checked and modified by agronomists, while offering clear advice to farmers, should be an interesting challenge.

2020 projects

• Automating Crop Canopy Data Collection for Crop Management
• Collecting Farm-sourced Data on Pest and Disease Pressure

NIAB is the UK’s leading Crop Science research centre. Within our portfolio of research we develop digital models to describe and predict crop development. This has significant implications for agriculture and more widely with regards availability of food.

We have a number of problems which I think would make interesting 1B projects and I was wondering what the process and timings are for submitting proposals?