Defra - Department for Environment, Food and Rural Affairs.

Science Search

Science and Research Projects

Return to Science Search homepage   Return to Project List

Machine vision for statutory seed certification. - VS0119T

Description
The EC Seeds Marketing Directives set standards for the number of seeds of other species in samples of certified seed. To test for conformity, a sample of prescribed weight, containing approximately 25,000 seeds is subject to the determination of other seeds by number test. This test is carried out by trained seed analysts who manually sift through samples and detect contaminants by eye. These are then removed by hand for identifiaction and counting. The task is both labour intensive and, if chemical seed treatments have been applied, presents a potential health hazard. Automation of this test could reduce the time required and minimise the health risk.

Various attempts have been made in recent years to automate detection and removal of contaminants from seed using machine vision technology. In such systems images of seeds are captured by a video camera and digitised using a microcomputer. Bespoke image analysis software is then used to extract seed features and data for both "sample" and "contaminant" seeds are used to train a classifier to distinguish between the two by a process called machine learning.

The purpose of this work is in three parts:
1. To develop a demonstration instrument which will automatically scan cereal samples, defective and diseased grain and other contaminants and sort them into appropriate categories according to The Cereal Seeds Regulations 1993 (Statutory Instrument ( S.I.) 199312005), as ammended. The concept will be based on previous work on image processing performed by the SASA and the Turing Institute and produce handling and presentation methodologies of which SAC has great experience. This represents a critical step towards a commercially viable system.

2. To utilise the demonstration instrument to develop the existing manually created data base. This facility will allow a quantum improvement in the ability to collect seed taxonometric statistics. There will also be an enhancement of the portfolio of proven handling and presentation techniques from potatoes and raspberries to objects the size of cereal grains. Techniques will include machine vision, image processing and multivariate statistical methods.

3. In the third year of the proposal, the demonstration instrument will be adapted for test under conditions for statutory seed testing and its performance assessed. These tests will be carried out in the laboratories of SASA (Official Seed Testing Station for Scotland). In addition, experiments will be carried under conditions for statutory seed testing in the laboratories of Alexander Harley
Objective
1. To develop a prototype semi-automatic grain/seed grading and sorting machine based on non-contact, machine vision-based technology already demonstrated by SASA and Turing Institute. The demonstrator system is essential to be able to examine a sufficient quantity of cereal to ensure statistical robustness when evaluating the system and any new developments thereafter.

2. To establish vision-based sensing methods for determining key quality assurance metrics from cereal samples according to The Cereal Seeds Regulations 1993 ( S.I. 1993/2005), as ammended that can be incorporated within the sorting machine, eg for the detection of the following within a sample:
• number of seed contaminants and their identification in 500 and 1 Kgm samples
• number of other contaminants and their type in 500 and 1 Kgm seed samples
• overall analytical purity on a percentage weight basis with other seed and inert matter contamination being identified
• on a weight basis with the other seed contaminants being identified

3. To collate a data base of cereal seed images (and their numeric descriptions) of the predominant UK species and varieties of these. This database would also contain examples of seed contaminants (weeds and other crop seeds).

4. To assess the performance of the sorting machine within typical operational parameters, particularly for statutory seed testing and classification the predominantly important cereal species, and varieties of these species, in the UK and the most prevalent types of contamination.
Project Documents
• Final Report : Machine vision for statutory seed certification   (608k)
Time-Scale and Cost
From: 2000

To: 2004

Cost: £163,523
Contractor / Funded Organisations
University - Scottish Agricultural College
Keywords
Certification              
Plant Varieties and Seeds              
Plants and Animals              
Quality              
SeedTesting              
Fields of Study
Plant Varieties and Seeds