NEC06665 Automatic image recognition for Japanese Knotweed for the conveyancing sector

Lead Research Organisation: NERC CEH (Up to 30.11.2019)
Department Name: Biodiversity (Wallingford)


We will address our current knowledge gaps with a programme of activities aimed at increasing our understanding of the market need and potential end-users, as well as producing a proof-of-concept image classifier.

Work to be undertaken:

Proof-of-concept algorithm - Using over 8,000 images of Japanese Knotweed and over 20,000 images of other common garden plants, we will produce a proof-of-concept image classification algorithm that can identify images of Japanese Knotweed, with an estimate of certainty, and can be integrated with the Japanese Knotweed risk map to provide an overall assessment of the likelihood of the species being present on a property. Building this proof-of-concept is key for follow-on funding as it will be the basis of any future application. Additionally, the limitations and capabilities of the technique are unknown until a classifier has been built and evaluated. A functional proof-of-concept should increase stakeholder recruitment and engagement in the follow-on application.

RICS consultation - We will consult with the Royal Institute of Chartered Surveyors (RICS) as the trade body for surveyors to discuss the needs gap in the industry, the size of the market and to set out the evaluation criteria of the image classifier. Understanding the market and the specific end-users will be key to the follow-on application. Through previous work with RICS, we know that they are experts in Japanese Knotweed surveying and remediation industry and they can contribute significantly to identifying markets for an image classification tool, as well as connecting us with end-users who could be involved in a follow-on application.

Classifier report - The project will collate images of Japanese Knotweed, primarily from the Biological Records Centre's system 'iRecord'. We will report on the availability of images, the impact of image and model selection on classifier accuracy, the computational requirements, and the accuracy that the classifiers are able to achieve in 'real-world' conditions. The technical limitations of the project, both in terms of images available, and the accuracy of the classifier are key for identifying specific end user applications. This understanding will be used to specify objectives for the follow-on application and to select end-users for the co-design phase.

Code base - We will produce a library of code for designing, training, and implementing image classifiers which will be the basis of the follow-on application. This code base will be the intellectual property of CEH, derived from CEH's data and image holdings, which will form the core of the follow-on application.

Planned Impact

Currently the value of images collected through iRecord is limited. The images are collected as a by-product of a requirement for evidence of what people record as seeing. These images allow experts to verify the recorders' sighting. These images do not currently have significant value beyond this use. The realisation of image classification technology in the past 5 years has dramatically increased the potential value of this wealth of expertly validated images. Without this pathfinder project, we cannot accurately estimate the true value of these images when used to create an image classifier. While we will focus on Japanese Knotweed in the pathfinder project, it is important to realise that the commercial implications of creating effective classifiers from our image database reach far beyond this one plant species. If the pathfinder is successful, it is hoped that the follow-on project can expand the scope of the classifier, for example identifying a range of garden weeds. This direction will be driven through consultation with RICS and other end-users.
A follow-on project will allow us to build a community of stakeholders to specify the best delivery mechanism for the service (e.g. smartphone application or website). The follow-on project will also seek to build upon the technical work of the pathfinder to improve the performance of the algorithm where opportunities are identified during the pathfinder project.


10 25 50
Description We have created an Artificial Intellegence, image classifier, which is able to identify images of Japanese Knotweed from images of other types of vegetation. This classifier has an accuracy of over 90%.

The key objectives of this grant have been met. This includes the production of a proof-of-concept classifier, consultation with potential users, scoping the availability of training images of Japanese Knotweed, and developing a code-base for future work.
Exploitation Route We hope to build on this work to create a tool to help surveyors and home-owners survey for Japanese Knotweed, as is required with the sale of every residential property.
Sectors Construction,Digital/Communication/Information Technologies (including Software),Environment,Financial Services, and Management Consultancy,Transport

Description Detecting the presence of invasive plant species: More quickly, cheaply and safely using AI and machine vision
Amount £90,136 (GBP)
Funding ID 48238 
Organisation Innovate UK 
Sector Public
Country United Kingdom
Start 04/2020 
End 02/2021
Description Improving Japanese Knotweed surveys using data and AI
Amount £20,230 (GBP)
Organisation UK Centre for Ecology & Hydrology 
Sector Public
Country United Kingdom
Start 05/2019 
End 04/2020
Description New Opportunities Fund
Amount £17,188 (GBP)
Funding ID NEC07118 
Organisation UK Centre for Ecology & Hydrology 
Sector Public
Country United Kingdom
Start 11/2018 
End 03/2020
Title Japanese Knotweed Image Classifier 
Description A machine learning algorithm that identifies images of the invasive plant species Japanese Knotweed, from images of other types of vegetation. 
Type Of Material Computer model/algorithm 
Year Produced 2018 
Provided To Others? No  
Impact None as yet, though this being used as the basis of proposals in preparation.