Final Report Summary - SPECIMFOOD (Spectral Imaging for Contaminant Detection on Fresh Food Produce)
The overall objective of this project was to develop and validate on-line, non-destructive spectral imaging technologies to rapidly assess safety and quality of fruits and vegetables. This will reduce food safety risks in pre-harvest and post-harvest production. This work focused primarily on fresh fruits and vegetables, such as leafy greens, that have been associated with serious agri-food safety related outbreaks, as well as other common products such as apples. The following were the specific objectives of this research:
Objective 1: Development and validation of multi-task in-line real-time inspection technologies for small to large processors which detect contaminants on fruits and vegetables.
Objective 2: Development and validation of portable spectral imaging sensing technologies for detecting the presence of contaminants on food products (in-field pre-harvest and during processing post-harvest) and on processing surfaces.
Firstly this project compared three hyperspectral imaging (HSI) configurations coupled with two multivariate image analysis techniques for detection of fecal contamination on spinach leaves (Spinacia oleracea). Hyperspectral fluorescence imaging using ultra violet (UV) and violet excitation sources, and hyperspectral reflectance imaging in the visible to near-infrared regions were investigated. Partial least squares discriminant analysis and two band ratio analysis techniques were used to assess and compare these HSI configurations. High detection accuracy was found for the two fluorescence HSI configurations compared to the visible/near infrared HSI. Both fluorescence HSI configurations had 100% detection rates for fecal contamination up to 1:10 dilution level, and violet HSI had 99% and 87% detection rates for 1:20 and 1:30 levels, respectively. Results indicated that fluorescence imaging with the violet excitation performed superior to HSI with UV excitation for detection of a range of diluted fecal contamination on leafy greens. 5% or less false positives were observed for the fluorescence HSI configurations, and they were associated with yellow hue on the leaves.
Hyperspectral fluorescence imaging with ultraviolet-A excitation was also used to evaluate the feasibility of two-waveband algorithms for the detection of bovine fecal contaminants on the abaxial and adaxial surfaces of Romaine lettuce. Correlation analysis was used to select the most significant waveband pairs for two band ratio and difference methods in distinguishing contaminated and uncontaminated leaf areas. Two-band ratios using bands at 665.6 nm and 680.0 nm were found to effectively differentiate all contamination spots applied to the lettuce. However, because the ratio images exhibited some false positives from leaf vein and inter-veinal regions, simple thresholding and morphological opening were also performed on the ratio images. The resultant binary images showed that all fecal contamination spots in the images for Romaine lettuce could be detected successfully without false positives.
Fluorescence emission profiles of spinach leaves were monitored over a 27 day storage period; peak emission blue-shifts were observed over the storage period accompanying a color change for green to green-yellow-brown hue. Violet fluorescence excitation was provided at 405 nm and light emission was recorded from 464 to 800 nm. Partial least square discriminant analysis and wavelength ratio methods were compared for detection accuracy for fecal contamination. A developed partial least squares discriminant analysis (PLSDA) model correctly detected fecal contamination on 100% of relatively fresh green spinach leaves used in this investigation, which also had soil contamination. A wavelength ratio technique using four wavebands (680, 688, 703 and 723 nm) was successful in identifying 100% of fecal contaminations on both fresh and non-fresh leaves.
The fluorescence emission peaks for fecal matter of animals that consume green plant materials and for chlorophyll a occur in close proximity in the red spectral region. Consequently, a high spectral resolution would be required for multispectral imaging for online implementation to detect bovine fecal contamination on leafy greens such as Romaine lettuce and baby spinach. An on-line fluorescence imaging inspection system for fecal contaminant detection has potential to allow fresh produce producers to reduce foodborne illnesses and prevent against the associated economic losses.
A multispectral algorithm for detection and differentiation of defective (defects on apple skin) and normal Red Delicious apples was developed from analysis of a series of hyperspectral line-scan images. A fast line-scan hyperspectral imaging system mounted on a conventional apple sorting machine was used to capture hyperspectral images of apples moving approximately 4 apples per second on a conveyor belt. The detection algorithm included an apple segmentation method and a threshold function, and was developed using three wavebands at 676 nm, 714 nm and 779 nm. The algorithm was executed on line-by-line image analysis, simulating online real-time line-scan imaging inspection during fruit processing. The rapid multispectral algorithm detected over 95% of defective apples and 91% of normal apples investigated. The algorithm could complete scanning and obtain the detection result for an apple in less than 0.03 seconds. The detection performance of the algorithm shows its potential for use with high-speed non-destructive machine vision systems to help ensure food safety and quality, increase efficiency, and reduce costs for the apple industry.
Food processing surfaces can act as a medium for cross-contamination if adequate sanitization procedures are not carried out. Ensuring food processing surfaces are correctly cleaned and sanitized is an important procedure in the food industry for reducing the risk of foodborne illnesses and related costs. A handheld fluorescence imaging device was engineered and assessed for detection of three types of food residues, which have been associated with foodborne illness outbreaks, on two commonly used processing surfaces, i.e. high-density polyethylene (HDPE) and food grade stainless steel (SS). The food residues assessed were from spinach leaf, fat free milk, and bovine red meat. Fluorescence excitation was supplied by 4 x 10 watt light emitting diodes at 405 nm. Interchangeable light filters were selected to optimize the contrast between the different food residues and processing surfaces, using hyperspectral fluorescence imaging. The handheld fluorescence imaging device along with image analysis distinguished the different food residues from the HDPE and SS processing surfaces. This optical sensing device can be used as a visual-aid for detection of food fouling on food processing surfaces over relatively large areas. This device has the potential to be used in the food industry as visual-aid for detection of specifically targeted food residues.
Detection of fecal contaminants on leafy greens in the field will allow for decreasing cross-contamination of produce during- and post-harvest. Fecal contamination of leafy greens has been associated with E.coli O157:H7 outbreaks and foodborne illnesses. Passive field spectroscopy, measuring reflectance and fluorescence created by the suns light, coupled with numerical normalization techniques we used to distinguish fecal contaminant on spinach leaves from soil on spinach leaves and uncontaminated spinach leaf samples. A Savitzky-Golay first derivative transformation and a waveband ratio of 710:688 nm as normalizing techniques were assessed. A SIMCA (soft independent modeling of class analogies) procedure with a 216 sample training set successfully predicted all 54 test set sample types, using the spectral region of 600-800 nm. The ratio of 710:688 nm along with set thresholds separated all 270 samples by type. Application of these techniques in-field to avoid harvesting of fecal contaminated leafy greens may lead to a reduction in produce waste by reducing opportunities for cross-contamination. These techniques may also reduce risks of foodborne illnesses.
Improving the safety and quality of produce consumed by the public is the key potential impact and use of these technologies. The systems developed can incorporate multitasking capabilities which can allow users to select desired inspection criteria, and to optimize wavelengths and thresholds to address changes in produce characteristics on-the-fly.
Contact Persons:
Prof. Colm O’Donnell: UCD School of Biosystems and Food Engineering, Agriculture and Food Science Centre, Belfield, Dublin 4, Ireland. Ph: +353-1-7167201. E-mail: colm.odonnell@ucd.ie
Dr. Colm Everard: UCD School of Biosystems and Food Engineering, Agriculture and Food Science Centre, Belfield, Dublin 4, Ireland. Ph: +353-1-7167463. E-mail: colm.everard@ucd.ie
Objective 1: Development and validation of multi-task in-line real-time inspection technologies for small to large processors which detect contaminants on fruits and vegetables.
Objective 2: Development and validation of portable spectral imaging sensing technologies for detecting the presence of contaminants on food products (in-field pre-harvest and during processing post-harvest) and on processing surfaces.
Firstly this project compared three hyperspectral imaging (HSI) configurations coupled with two multivariate image analysis techniques for detection of fecal contamination on spinach leaves (Spinacia oleracea). Hyperspectral fluorescence imaging using ultra violet (UV) and violet excitation sources, and hyperspectral reflectance imaging in the visible to near-infrared regions were investigated. Partial least squares discriminant analysis and two band ratio analysis techniques were used to assess and compare these HSI configurations. High detection accuracy was found for the two fluorescence HSI configurations compared to the visible/near infrared HSI. Both fluorescence HSI configurations had 100% detection rates for fecal contamination up to 1:10 dilution level, and violet HSI had 99% and 87% detection rates for 1:20 and 1:30 levels, respectively. Results indicated that fluorescence imaging with the violet excitation performed superior to HSI with UV excitation for detection of a range of diluted fecal contamination on leafy greens. 5% or less false positives were observed for the fluorescence HSI configurations, and they were associated with yellow hue on the leaves.
Hyperspectral fluorescence imaging with ultraviolet-A excitation was also used to evaluate the feasibility of two-waveband algorithms for the detection of bovine fecal contaminants on the abaxial and adaxial surfaces of Romaine lettuce. Correlation analysis was used to select the most significant waveband pairs for two band ratio and difference methods in distinguishing contaminated and uncontaminated leaf areas. Two-band ratios using bands at 665.6 nm and 680.0 nm were found to effectively differentiate all contamination spots applied to the lettuce. However, because the ratio images exhibited some false positives from leaf vein and inter-veinal regions, simple thresholding and morphological opening were also performed on the ratio images. The resultant binary images showed that all fecal contamination spots in the images for Romaine lettuce could be detected successfully without false positives.
Fluorescence emission profiles of spinach leaves were monitored over a 27 day storage period; peak emission blue-shifts were observed over the storage period accompanying a color change for green to green-yellow-brown hue. Violet fluorescence excitation was provided at 405 nm and light emission was recorded from 464 to 800 nm. Partial least square discriminant analysis and wavelength ratio methods were compared for detection accuracy for fecal contamination. A developed partial least squares discriminant analysis (PLSDA) model correctly detected fecal contamination on 100% of relatively fresh green spinach leaves used in this investigation, which also had soil contamination. A wavelength ratio technique using four wavebands (680, 688, 703 and 723 nm) was successful in identifying 100% of fecal contaminations on both fresh and non-fresh leaves.
The fluorescence emission peaks for fecal matter of animals that consume green plant materials and for chlorophyll a occur in close proximity in the red spectral region. Consequently, a high spectral resolution would be required for multispectral imaging for online implementation to detect bovine fecal contamination on leafy greens such as Romaine lettuce and baby spinach. An on-line fluorescence imaging inspection system for fecal contaminant detection has potential to allow fresh produce producers to reduce foodborne illnesses and prevent against the associated economic losses.
A multispectral algorithm for detection and differentiation of defective (defects on apple skin) and normal Red Delicious apples was developed from analysis of a series of hyperspectral line-scan images. A fast line-scan hyperspectral imaging system mounted on a conventional apple sorting machine was used to capture hyperspectral images of apples moving approximately 4 apples per second on a conveyor belt. The detection algorithm included an apple segmentation method and a threshold function, and was developed using three wavebands at 676 nm, 714 nm and 779 nm. The algorithm was executed on line-by-line image analysis, simulating online real-time line-scan imaging inspection during fruit processing. The rapid multispectral algorithm detected over 95% of defective apples and 91% of normal apples investigated. The algorithm could complete scanning and obtain the detection result for an apple in less than 0.03 seconds. The detection performance of the algorithm shows its potential for use with high-speed non-destructive machine vision systems to help ensure food safety and quality, increase efficiency, and reduce costs for the apple industry.
Food processing surfaces can act as a medium for cross-contamination if adequate sanitization procedures are not carried out. Ensuring food processing surfaces are correctly cleaned and sanitized is an important procedure in the food industry for reducing the risk of foodborne illnesses and related costs. A handheld fluorescence imaging device was engineered and assessed for detection of three types of food residues, which have been associated with foodborne illness outbreaks, on two commonly used processing surfaces, i.e. high-density polyethylene (HDPE) and food grade stainless steel (SS). The food residues assessed were from spinach leaf, fat free milk, and bovine red meat. Fluorescence excitation was supplied by 4 x 10 watt light emitting diodes at 405 nm. Interchangeable light filters were selected to optimize the contrast between the different food residues and processing surfaces, using hyperspectral fluorescence imaging. The handheld fluorescence imaging device along with image analysis distinguished the different food residues from the HDPE and SS processing surfaces. This optical sensing device can be used as a visual-aid for detection of food fouling on food processing surfaces over relatively large areas. This device has the potential to be used in the food industry as visual-aid for detection of specifically targeted food residues.
Detection of fecal contaminants on leafy greens in the field will allow for decreasing cross-contamination of produce during- and post-harvest. Fecal contamination of leafy greens has been associated with E.coli O157:H7 outbreaks and foodborne illnesses. Passive field spectroscopy, measuring reflectance and fluorescence created by the suns light, coupled with numerical normalization techniques we used to distinguish fecal contaminant on spinach leaves from soil on spinach leaves and uncontaminated spinach leaf samples. A Savitzky-Golay first derivative transformation and a waveband ratio of 710:688 nm as normalizing techniques were assessed. A SIMCA (soft independent modeling of class analogies) procedure with a 216 sample training set successfully predicted all 54 test set sample types, using the spectral region of 600-800 nm. The ratio of 710:688 nm along with set thresholds separated all 270 samples by type. Application of these techniques in-field to avoid harvesting of fecal contaminated leafy greens may lead to a reduction in produce waste by reducing opportunities for cross-contamination. These techniques may also reduce risks of foodborne illnesses.
Improving the safety and quality of produce consumed by the public is the key potential impact and use of these technologies. The systems developed can incorporate multitasking capabilities which can allow users to select desired inspection criteria, and to optimize wavelengths and thresholds to address changes in produce characteristics on-the-fly.
Contact Persons:
Prof. Colm O’Donnell: UCD School of Biosystems and Food Engineering, Agriculture and Food Science Centre, Belfield, Dublin 4, Ireland. Ph: +353-1-7167201. E-mail: colm.odonnell@ucd.ie
Dr. Colm Everard: UCD School of Biosystems and Food Engineering, Agriculture and Food Science Centre, Belfield, Dublin 4, Ireland. Ph: +353-1-7167463. E-mail: colm.everard@ucd.ie