Search

Hyperspectral (HS) images have the characteristic of containing both spectral and spatial information of a sample. Typically, spectral information can be related with chemical and physical properties through multivariate regression models. The application of these models onto HS images results in prediction maps, which provide an estimation of the modelled chemical information for each pixel of the image. This approach has wide applications in food processing industries for online monitoring of product quality and process control. One of the main difficulties derived from an imaging set up, is that the size of the pixels is usually much smaller than the area required to obtain a wet chemical reference. This means that, as opposed to point spectroscopy, the performance of the estimations cannot be evaluated by directly comparing observed and estimated values for each pixel. Moreover, the selection of regression model parameters, such as the number of latent variables (LV) in a partial least squares (PLS) model, cannot be assessed on a pixel basis either. Nonetheless, compared to point spectroscopy, HS imaging does provide information on the spatial distribution of the predicted values. The objective of this work is to propose a quantitative approach to use spatial information of prediction maps for supporting the evaluation of regression models applied to HS images. This approach is based on the use of geostatistical indexes, which allow decomposing the total variance of the prediction maps into two components: non spatially structured and spatially structured variance, represented respectively by the nugget effect (C0) and the partial sill (C1). This strategy was tested in a simulated dataset and two real case studies. Geostatistical indices of the prediction maps were compared to model performance metrics for PLS models with increasing number of LV. As a result, this work stablishes a connection between linear regression model performance estimates and the spatial decomposition of variance in prediction maps, when the ground truth to estimate is spatially structured. The presented study [1], allowed to evaluate HS imaging models, not only from average estimations which can be compared to reference values, but also from the spatial structure of prediction maps. This approach does not require ground truth values and could be used as a source of information for supporting the choice of optimum calibration options, such as the number of LV, or the pre-treatments, complementing the traditional visual inspection of prediction maps with quantitative and objective metrics. Further works could explore the effect of spatial structure at different scales in the data. This approach could also be applied to evaluate variations of other parameters in the model, such as different methods for spectral pretreatments.

Pushing HSI to the Limit for the Quantification of Peanut Traces in Bulk Powder Foods

NIR Hyperspectral imaging (1000-2500 nm) combined with IDC allowed the detection of peanut traces down to adulteration percentages 0.01% Contrary to PLSR, IDC does not require a calibration set, but uses both expert and experimental information and suitable for quantification of an interest compound in complex matrices. The obtained results shows the feasibility of using HSI systems for the detection of peanut traces in conjunction with chemical procedures, such as RT-PCR and ELISA

NIR Hyperspectral images (1000-2200 nm) allowed the detection of peanut traces down to adulteration percentages 0.01 % - Determination coefficient of R2= 0.946 was found for the quantification of peanut adulteration from 10% to 0.1%. - The obtained results shows the feasibility of using HSI systems for the detection of peanut traces in conjuction with chemical procedures, such as RT-PCR and ELISA to facilitate quality control surveyance on food product processing lines.

Comunicaciones presentadas en el marco del proyecto FRUTURA de CYTED. Reunión Científico Técnica FRUTURA San José-EARTH (Costa Rica) - Mayo de 2011

Background: Analysis of exhaled volatile organic compounds (VOCs) in breath is an emerging approach for cancer diagnosis, but little is known about its potential use as a biomarker for colorectal cancer (CRC). We investigated whether a combination of VOCs could distinct CRC patients from healthy volunteers. Methods: In a pilot study, we prospectively analyzed breath exhalations of 38 CRC patient and 43 healthy controls all scheduled for colonoscopy, older than 50 in the average-risk category. The samples were ionized and analyzed using a Secondary ElectroSpray Ionization (SESI) coupled with a Time-of-Flight Mass Spectrometer (SESI-MS). After a minimum of 2 hours fasting, volunteers deeply exhaled into the system. Each test requires three soft exhalations and takes less than ten minutes. No breath condensate or collection are required and VOCs masses are detected in real time, also allowing for a spirometric profile to be analyzed along with the VOCs. A new sampling system precludes ambient air from entering the system, so background contamination is reduced by an overall factor of ten. Potential confounding variables from the patient or the environment that could interfere with results were analyzed. Results: 255 VOCs, with masses ranging from 30 to 431 Dalton have been identified in the exhaled breath. Using a classification technique based on the ROC curve for each VOC, a set of 9 biomarkers discriminating the presence of CRC from healthy volunteers was obtained, showing an average recognition rate of 81.94%, a sensitivity of 87.04% and specificity of 76.85%. Conclusions: A combination of cualitative and cuantitative analysis of VOCs in the exhaled breath could be a powerful diagnostic tool for average-risk CRC population. These results should be taken with precaution, as many endogenous or exogenous contaminants could interfere as confounding variables. On-line analysis with SESI-MS is less time-consuming and doesn’t need sample preparation. We are recruiting in a new pilot study including breath cleaning procedures and spirometric analysis incorporated into the postprocessing algorithms, to better control for confounding variables.

The production of minimally processed vegetables and fruits is an emergent sector, however these processes reduce the useful life of the products. Main preservation techniques such cold storage and modified atmosphere are limited. New treatments are being applied (O3 , UV‐C radiation, biodegradable films…etc.). The sector precise of cheap and fast techniques to evaluate the general quality and the security of the processed products, that constitute a tool of aid to the decision in the implementation of new procedures of packaging and/or treatments. Objectives: To explore hyperspectral imaging for monitoring the evolution of minimally processed leafy vegetables during shelf‐life . To identify and classify deterioration rates of the leaves through Multivariate analysis techniques (PLS‐DA)

La presente investigación se centra en el uso de la visión artificial para determinar la maduración del melocotón rojo de pulpa blanda (`Richlady'). La visión artificial permite una determinación espacial detallada del estado de maduración del fruto. Los índices ópticos considerados (Ind1 e Ind2, propuestos en la actual investigación, e Ind3 e IAD, propuestos por otros autores) se basan en la combinación de longitudes de onda en la zona del pico de absorción de la clorofila (680 nm). Ind1 corresponde aproximadamente a la profundidad del pico de absorción, e Ind2 corresponde a la profundidad relativa. Se obtuvo una imagen artificial de cada índice computada a partir de imágenes hiperespectales. Todos los índices fueron capaces de corregir el efecto de la convexidad (a excepción de los melocotones recién cosechados y para el Ind1). Ind2 es el índice con mayor capacidad de discriminación en diferentes estados de madurez. Por otra parte Ind2 permite la diferenciación de regiones de maduración dentro de los frutos, mostrando la evolución de esas regiones durante la maduración. The present research is focused on the application of artificial vision to assess the ripening of red skinned softflesh peach (‘Richlady’). Artificial vision allows a spatially detailed determination of the ripening stage of the fruit. The considered optical indexes (Ind1 and Ind2, proposed in the present research, and Ind3 and IAD, proposed by other authors) are based on the combination of wavelengths close to the chlorophyll absorption peak at 680 nm. Ind1 corresponds approximately to the depth of the absorption peak, and Ind2 corresponds to the relative absorption peak. An artificial image of each index was obtained by computing the corresponding reflectance images, which were acquired with a hyperspectral camera. All indexes were able to correct convexity (except for the just-harvested peaches and for Ind1). Ind2 is the preferred index; it showed the highest discriminating power between ripening stages and no influence of convexity. Ind2 also allowed the differentiation of ripening regions within the fruits, and it showed the evolution of those regions during ripening.

Resumen En el presente trabajo se explora la técnica de imagen hiperespectral en el ámbito de las hortalizas de hoja mínimamente procesadas para la determinación de atributos de calidad ligados a la evolución durante su almacenamiento y manejo. Se ha implementado un equipo de visión hiperespectral VIS-NIR (400 – 1000 nm) para la adquisición de imágenes de hojas de espinacas. Las muestras han sido sometidas a diferentes periodos de almacenamiento para generar suficiente variabilidad en estados de calidad. Se ha seleccionado una población de calibración de espectros sobre las imágenes considerando tres categorías de calidad. Sobre dicha población de calibración se ha aplicado la técnica de análisis multivariante PLS-DA. El error del modelo de clasificación en la calibración ha sido del 7%. La proyección de las imágenes hiperespectrales en el espacio discriminante generado y la asignación de cada píxel a una de las categorías en función de dicha proyección, han permitido identificar en las hojas regiones con diferentes estados de evolución. Abstract In this paper hyperspectral imaging technique is explored for the determination of quality attributes related to the evolution during storage and handling in the field of minimally processed leafy vegetables. We have implemented a computer vision hyperspectral VIS-NIR (400 - 1000 nm) for the acquisition of images of leaves of spinach. Samples were subjected to different storage periods to generate sufficient variability in quality stages. It has been selected a calibration set of spectra on the images by considering three categories of quality. On this calibration set the multivariate PLS-DA has been performed. The classification error in calibration was 7%. The projection of hyperspectral images on the discriminant space generated and the assignation of each pixel to one of the categories, have allowed the identification of regions in the leaves with different stages of evolution.

Diversas investigaciones han intentado resolver el problema de identificación de frutos u hojas mediante imágenes digitales, pero sólo lo han logrado parcialmente. Por esto, el objetivo de este trabajo es explorar una metodología de identificación que permita estimar áreas de hojas y racimos en viñedos, empleando imágenes en el espectro visible (RGB) y en el infrarrojo cercano (RGIR). El problema de la identificación fue abordando por dos vías, forma y color. En el caso de la identificación por forma se empleó la transformada circular de Hough y en el de la identificación por color se emplearon las técnicas de clasificación no supervisada denominadas kmeans y Fuzzy c-means. Se determinó que la clasificación mediante k-means sobre el espacio L*a*b*, para imágenes RGB y sobre el índice SAVI en las imágenes RGIR, son las técnicas más adecuadas. En cuanto a la identificación por forma, ésta resultó aplicable sólo en condiciones muy particulares

Vine vigour and fruit-cluster exposure to sunlight in a grapevine canopy fruiting zone has been shown to strongly correlate with key fruit composition and diseases incidence. In this framework, the use of automated image analysis for the identification of plant elements is an important issue to be addressed for vineyard assessment (Dunn and Martin, 2004). In addition, optimum segmentation method is strongly application dependent and thus needs to be tested for each particular case (Cheng et al., 2001). The objective of the present work is to propose and test a simple, rapid and practical method for the identification of two relevant elements of grapevines canopy: clusters and green leaves.

Three groups of measurements related to peach maturity were acquired through destructive (D) mechanical tests (Magness Taylor Firmness, MTF), mechanical non destructive (ND) tests, and ND optical spectroscopy (Optical indexes). The relationship between these groups of variables was studied in order to estimate D mechanical measurements (MTF, with higher instrumental and sampling variability, time consuming, generally used as a reference for the assessment of peach handling), from ND measurements (quick, applicable on line, dealing better with the high variability found in fruit products). Multivariate exploratory analysis was used to extract the structure of the data. The information about the data structure of ND measurements, the relationship of MTF with the space defined by ND variables, and the expert knowledge regarding to the dataset was then used for modelling MTF (R 2 =0.72 and standard error on validation 5.73 N)

The present research is focused on the application of artificial vision to assess the ripening of red skinned soft-flesh peach (Richlady'). Artificial vision allows a spatially detailed determination of the ripening stage of the fruit. The considered optical indexes (Ind1 and Ind2, proposed in the present research, and Ind3 and IAD, proposed by other authors) are based on the combination of wavelengths close to the chlorophyll absorption peak at 680 nm. Ind1 corresponds approximately to the depth of the absorption peak, and Ind2 corresponds to the relative absorption peak. An artificial image of each index was obtained by computing the corresponding reflectance images, which were acquired with a hyperspectral camera. All indexes were able to correct convexity (except for the just-harvested peaches and for Ind1). Ind2 is the preferred index; it showed the highest discriminating power between ripening stages and no influence of convexity. Ind2 also allowed the differentiation of ripening regions within the fruits, and it showed the evolution of those regions during ripening.

El presente trabajo presenta diferentes técnicas no destructivas para la determinación de parámetros de calidad en frutas y hortalizas. El desarrollo de estas técnicas para la selección de un elevado número de individuos, y su posible implementación en equipos, tanto de laboratorio, como portátiles que permitan la medición directa en campo, supone una valiosa herramienta para el mejorador. La rápida determinación de parámetros de calidad para un elevado número de individuos resulta de gran ayuda en programas de mejora. Se presentan ejemplos de distintas aplicaciones: Equipos y procedimientos NIR en la mejora del contenido en sólidos solubles (SS) en cebollas; espectrometría en reflectancia y en imagen para establecer la madurez en recolección y en posrecolección en melocotón; y unas primeras aplicaciones de metabonómica para el estudio de aceitunas individuales, procedentes de cruzamientos de distintos cultivares.

The present research is focused on the application of artificial vision to peach ripening assessment, avoiding multiplicative and additive effect. Original images were acquired with a hyperspectral camera. Vision allows a spatially detailed determination of the ripening stage of the fruit. Optical indexes are proposed, based on the combination of wavelengths close to the chlorophyll absorption peak at 680 nm. Ind1 corresponds approximately to the depth of the absorption peak, and Ind2 corresponds to the relative absorption peak. An artificial image of each index was obtained by computing the corresponding reflectance images. Score images have been also computed from Principal Components and Partial Least Squares Analysis. In any case the best performances correspond to such images that correct multiplicative and additive effects. Ind2 is the preferred index; it showed the highest discriminating power between ripening stages and no influence of convexity. Ind2 also allowed the differentiation of ripening regions within the fruits, and it showed the evolution of those regions during ripening. This fact has been also observed in some of the score images

La demanda por parte del consumidor de frutos con sabo ry textura garantizados hace que los productores, centrales de confeccion y comercializadores busquen herramientas que les permitan conocer la calidad interna del producto que manipulan, para satisfacer esa demanda de forma consistente. En este articulo se presentan los resultados preliminares obtenidos con distintos equipos de estimacion no destructiva de la calidad interna en comparacion con los metodos tradicionales de medida.

Non destructive determination of stone fruit firmness is a critical factor for improving its quality and handling. This work applies available acoustic techniques to the determination peach firmness, considering and comparing different parameters obtained with 2 devices: AWETA-AFS (Acoustic Firmness Sensor), which integrates impact and acoustic response information; and a prototype device by LPF-TAG, which enables to analyze the whole acoustic spectra. Magness-Taylor, quasiestatic ball compression and non-destructive impact are used as firmness references, obtaining up to 86% R. La determinación no destructiva de la firmeza en melocotón es fundamental para mejorar la calidad y facilitar el manejo de esta fruta por la industria. El presente trabajo se plantea con el objetivo de estudiar la aplicación de técnicas acústicas en la caracterización de firmeza en melocotón. Para ello se evalúan y comparan distintos parámetros obtenidos mediante un sensor comercial, el dispositivo “Acoustic Firmness Sensor” (ASF de AWETA) que incorpora también un sensor de impacto; y un prototipo de equipo acústico desarrollado en el LPF-TAG, que permite analizar el espectro acústico completo. Como medidas de referencia se utilizaron: el ensayo Magness-Taylor (MT), la compresión quasiestática con contacto esférico (B) y la respuesta al impacto, ensayo no destructivo (IMP.) Como principales resultados destacan las correlaciones entre algunos de los parámetros acústicos con las variables de referencia (R de hasta 0,86)