Towards the automation of food quality/contamination assessment via non invasive techniques

Transcription

1 Towards the automation of food quality/contamination assessment via non invasive techniques Tsakanikas P., Panagou E. Z., Nychas G.-J. E. {p.tsakanikas, stathispanagou, P. Tsakanikas, IFPAC2015 1

2 Current Problem Introduction OverView Methodology Identification/Characterization Evaluation of Procedure Spectra Conclusions Results & Ongoing & Conclusions Work P. Tsakanikas IFPAC2015 2

3 Future Aim: Create a system that will enable monitoring the quality of meat from the start to the consumers plate. P. Tsakanikas IFPAC2015 3

4 Why? Ethical: The food policy must be built around high food safety standards, which serve to protect, and promote, the health of the consumer. Food Industries are primarily RESPONSIBLE for Food Safety. Consumers SHOULD be offered a wide range of SAFE and High Quality products. Practical (i.e. Economical): Food Industries will benefit: Better quality assessment on ALL of their products. Faster and direct quality control. Gain feedback for their production and distribution chain. Gain/enjoy the TRUST of the consumers. Our Goal: Create a user-free automated system for Food Quality assessment!!! P. Tsakanikas IFPAC2015 4

5 Vision! Identification and Characterization of Signature spectra from Samples VideoMeterLab Images Image Segmentation/ /ROI Detection Pixel spectra extraction Pre-Clustering of Spectra Relate the classes to microbial counts Biological meaning based Assignment of classes Spectral Clustering Spatial assignment of Contamination i.e. Contamination Maps, Signature database for online systems. 5

6 MultiSpectral Imaging Multispectral imaging Device VideometerLab 18 wavelengths from 405nm to 970 (NIR), not uniformly distributed P. Tsakanikas IFPAC2015 4

7 Automated Image Processing Develope Image Processing Workflow for useful sample detection and extraction: Fully automated and totally depended on the data. Free the data and subsequent analysis from human errors and biases. Status Information Display User Controls Input Image Segmented Image Display 7

8 Automated Image Processing (workflow) Informative Area Extraction Fat Exclusion Background/Surrounding Exclusion 6

9 Automated Image Processing (examples) 9

10 Automated Image Processing (examples) Submitted to: Food and Bioprocess Technology: An International Journal, Springer, under Review 10

11 Vision! Identification and Characterization of Signature spectra from Samples VideoMeterLab Images Image Segmentation/ /ROI Detection Pixel spectra extraction Pre-Clustering of Spectra Relate the classes to microbial counts Biological meaning based Assignment of classes Spectral Clustering Spatial assignment of Contamination i.e. Contamination Maps, Signature database for online systems. 11

12 Spectra Pre-Clustering ~ spectra one for each pixel!! A pre-clustering scheme is needed in order to reduce data!!!

13 Spectra Pre-Clustering K-means clustering to 100 final spectral classes (as the reflectance values 0-100%) Standard Normal Variate Normalization of spectra! Reduce the correlated information among wavelengths

14 Remap pre-clustering to image Assign pre-clustering labels to an original-like image -> clusters topologies Neighborhood pixels do not belong to the same class and same class is not located in just one site on the sample Image showing the topologies of the 100 clusters. Areas with the same label-color represent areas that have the same spectral signature

15 Unsupervised Clustering Identification of Natural classes Cluster Cluster

16 Remap clusters to image AF10 2.hips Cluster Cluster

17 System OverView Experimental Design 20 Sterile pork fillet 10 o C ~2log TVC SNV Pre-clustering Clustering Contaminated pork fillet 10 o C ~3 to 9log TVC SNV Pre-clustering Clustering P. Tsakanikas IFPAC

18 Results on images 2B10.hips 8AB10.hips 16AB10.hips ~2 TVC ~2 TVC ~2 TVC P10.hips Storage Time + TVC increment in Non sterile 5P10.hips 10AP10 1.hips ~2.9 TVC ~4.5 TVC 200 ~5.7 TVC

19 Conclusions The proposed method can be considered a novelty towards non-invasive quality assessment of food products, filling the gap in the area of food science concerning the unbiased information extraction from surface chemistry sensors. The investigation of those spectra and spatial distribution can be used as a pointer to presence of possible contamination. Once the pipeline is fully validated(!), then it should be straightforward its application to other types of food (maybe easier in some cases) AND OTHER TYPES OF SPECTRA! Innovative Workflow in Hyper/Multi-spectral processing with great impact!!! Future pointers Creation of 3-D multi spectro temporal signature for direct assignment of spectra to time and, through image spatial information, to space Food Quality specific Fingerprint!!! 19

20 Acknowledgments AUA team Nychas G.J. Panagou E.Z. Skandamis P.N Loukas D. Blana V. Ropodi A. Pavlidis D. The project "Intelligent multi-sensor system for meat analysis - imeatsense_550" has been cofinanced by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program: ARISTEIA-I. 20

21 Questions Thank you! Any Questions? P. Tsakanikas IFPAC