High Speed Single Kernel Imaging
(2015) In Master’s Theses in Mathematical Sciences FMA820 20132Mathematics (Faculty of Engineering)
- Abstract
- Food safety and quality has been an emerging trend on the food market in recent years. In the grain and cereals market focus is shifting from looking at grain as a bulk commodity to look at it as individual kernels. In the pursuit of quality improvement and assurance various methods are emerging attempting to sort grain, all grain, according to quality on a single kernel basis. One such technology is the TriQ technology developed by BoMill AB that by means of near infrared transmittance (NIT) spectroscopy can assess and subsequently sort individual kernels of grain by quality on a commercial scale. The quality traits currently available to sort by are protein content in soft wheat, durum wheat and malting barley, remove fusarium-infected... (More)
- Food safety and quality has been an emerging trend on the food market in recent years. In the grain and cereals market focus is shifting from looking at grain as a bulk commodity to look at it as individual kernels. In the pursuit of quality improvement and assurance various methods are emerging attempting to sort grain, all grain, according to quality on a single kernel basis. One such technology is the TriQ technology developed by BoMill AB that by means of near infrared transmittance (NIT) spectroscopy can assess and subsequently sort individual kernels of grain by quality on a commercial scale. The quality traits currently available to sort by are protein content in soft wheat, durum wheat and malting barley, remove fusarium-infected kernels from soft wheat, durum wheat and malting barley and to sort durum wheat by vitreousness.
In a step to further advance the TriQ technology this thesis aims to introduce an imaging system into the setup in order to evaluate the possibilities and potential benefits of real time image analysis in conjunction with the current NIT system.
For this purpose a crude camera setup was established to film the kernels during sorting and a software system was developed to, post sorting, merge sorting data with the corresponding kernel images and manually classify them by means of position properties that are of high relevance for the NIT measurement. This data is then used as ground truth for training of image classifiers to automatically classify the kernels. The software system also includes a system for benchmarking of different classifiers.
Poor image quality limited the classifier performance to an overall error rate of 25% for classification into five mutually exclusive position categories. Despite the relatively poor result the system has shown staggering potential for generating new sorting applications as well as being used as an evaluation and development tool for key parts in the TriQ technology along with providing a grading system for the quality distribution of the raw material being sorted.
Suggestions for future work would involve adopting the imaging setup as a tool for development of sorting algorithms for the current TriQ technology and to, based on the experiences from this thesis, build a proper imaging setup able to perform real time image analysis during sorting as a step closer towards a commercialization of this combined image- and spectra sorting technology. (Less) - Popular Abstract
- 25’000 single kernels of grain per second can be individually assessed for quality and sorted accordingly by BoMill’s TriQ technology. This corresponds to a capacity of 3 metric tonnes per hour. The technology uses near infrared transmittance (NIT) spectrometry to assess the chemical composition and internal structure of individual kernels of grain and is one of many emerging technologies in the food industry to meet the increasing demand for food quality and safety.
Previous technologies in the grain industry use computer vision for quality assessment so in a step to further advance the TriQ technology an imaging system was introduced into the sorting setup of a small scale unit, based on the TriQ technology, in order to evaluate the... (More) - 25’000 single kernels of grain per second can be individually assessed for quality and sorted accordingly by BoMill’s TriQ technology. This corresponds to a capacity of 3 metric tonnes per hour. The technology uses near infrared transmittance (NIT) spectrometry to assess the chemical composition and internal structure of individual kernels of grain and is one of many emerging technologies in the food industry to meet the increasing demand for food quality and safety.
Previous technologies in the grain industry use computer vision for quality assessment so in a step to further advance the TriQ technology an imaging system was introduced into the sorting setup of a small scale unit, based on the TriQ technology, in order to evaluate the possibilities and potential benefits of computer vision in conjunction with the current NIT system. The kernels were recorded during sorting and a software system was developed to, post sorting, merge sorting data with the corresponding kernel images and to manually classify them by means of position properties that are of high relevance for the NIT measurement.
This data was then used as ground truth for training of image classifiers to automatically classify the kernels. The software system also includes a system for benchmarking of different classifiers. In total five different position categories were used, chosen based on their impact on the NIT spectra.
Poor image quality limited the classifier performance to an overall error rate of 25% for classification into the five mutually exclusive position categories. Despite the relatively poor result the system has shown great potential for generating new sorting applications as well as being used as an evaluation and development tool for key parts in the TriQ technology along with providing a grading system for the quality distribution of the grain being sorted.
Combining the two most advanced technologies for grain processing available poses some severe challenges but as the thesis shows the potential benefits of such a merger are staggering. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/5436216
- author
- Lyko Biedermann, Hansi LU
- supervisor
-
- Karl Åström LU
- organization
- course
- FMA820 20132
- year
- 2015
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Grain, Cereal, Spectrometry, Image analysis
- publication/series
- Master’s Theses in Mathematical Sciences
- report number
- LUTFMA-3276-2015
- ISSN
- 1404-6342
- other publication id
- 2015:E16
- language
- English
- id
- 5436216
- date added to LUP
- 2015-06-18 12:10:27
- date last changed
- 2015-06-18 12:10:27
@misc{5436216,
abstract = {{Food safety and quality has been an emerging trend on the food market in recent years. In the grain and cereals market focus is shifting from looking at grain as a bulk commodity to look at it as individual kernels. In the pursuit of quality improvement and assurance various methods are emerging attempting to sort grain, all grain, according to quality on a single kernel basis. One such technology is the TriQ technology developed by BoMill AB that by means of near infrared transmittance (NIT) spectroscopy can assess and subsequently sort individual kernels of grain by quality on a commercial scale. The quality traits currently available to sort by are protein content in soft wheat, durum wheat and malting barley, remove fusarium-infected kernels from soft wheat, durum wheat and malting barley and to sort durum wheat by vitreousness.
In a step to further advance the TriQ technology this thesis aims to introduce an imaging system into the setup in order to evaluate the possibilities and potential benefits of real time image analysis in conjunction with the current NIT system.
For this purpose a crude camera setup was established to film the kernels during sorting and a software system was developed to, post sorting, merge sorting data with the corresponding kernel images and manually classify them by means of position properties that are of high relevance for the NIT measurement. This data is then used as ground truth for training of image classifiers to automatically classify the kernels. The software system also includes a system for benchmarking of different classifiers.
Poor image quality limited the classifier performance to an overall error rate of 25% for classification into five mutually exclusive position categories. Despite the relatively poor result the system has shown staggering potential for generating new sorting applications as well as being used as an evaluation and development tool for key parts in the TriQ technology along with providing a grading system for the quality distribution of the raw material being sorted.
Suggestions for future work would involve adopting the imaging setup as a tool for development of sorting algorithms for the current TriQ technology and to, based on the experiences from this thesis, build a proper imaging setup able to perform real time image analysis during sorting as a step closer towards a commercialization of this combined image- and spectra sorting technology.}},
author = {{Lyko Biedermann, Hansi}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master’s Theses in Mathematical Sciences}},
title = {{High Speed Single Kernel Imaging}},
year = {{2015}},
}