Wheat grain quality is defined by several parameters, of which insect and fungal damage and sprouting are considered important degrading factors. At present, Canadian wheat is inspected and graded manually by Canadian Grain Commission (CGC) inspectors at grain handling facilities or in the CGC laboratories. Visual inspection methods are time consuming, less efficient, subjective, and require experienced personnel. Therefore, an alternative, rapid, objective, accurate, and cost effective technique is needed for grain quality monitoring in real-time which can potentially assist or replace the manual inspection process. Insect-damaged wheat samples by the species of rice weevil (Sitophilus oryzae), lesser grain borer (Rhyzopertha dominica), rusty grain beetle (Cryptolestes ferrugineus), and red flour beetle (Tribolium castaneum); fungal-damaged wheat samples by the species of storage fungi namely Penicillium spp., Aspergillus glaucus, and Aspergillus niger; and artificially sprouted wheat kernels were obtained from the Cereal Research Centre (CRC), Agriculture and Agri-Food Canada, Winnipeg, Canada. Field damaged sprouted (midge-damaged) wheat kernels were procured from five growing locations across western Canada. Healthy and damaged wheat kernels were imaged using a long-wave near-infrared (LWNIR) and a short-wave near-infrared (SWNIR) hypersprctral imaging systems and an area scan color camera. The acquired images were stored for processing, feature extraction, and algorithm development. The LWNIR classified 85-100% healthy and insect-damaged, 95-100% healthy and fungal-infected, and 85-100% healthy and sprouted/midge-damaged kernels. The SWNIR classified 92.7-100%, 96-100% and 93.3-98.7% insect, fungal, and midge-damaged kernels, respectively (up to 28% false positive error). Color imaging correctly classified 93.7-99.3%, 98-100% and 94-99.7% insect, fungal, and midge-damaged kernels, respectively (up to 26% false positive error). Combined the SWNIR features with top color image features correctly classified 91-100%, 99-100% and 95-99.3% insect, fungal, and midge- damaged kernels, respectively with only less than 4% false positive error.

The agriculture industry experiences severe economic losses each year when wheat crops become infected with Fusarium and the mycotoxin Deoxynivalenol (DON). This research investigated the feasibility of using near infrared hyperspectral imaging to detect Fusarium damage and DON in Canadian Western Red Spring wheat. Four samples were selected from each grain grade resulting in 16 samples and 240 hyperspectral data cubes. The data cubes were calibrated to the system, the consistent spectra was found and then a 1- nearest neighbour classifier was generated. Grade percentages were computed and used to generate two 3- nearest neighbour classifiers, one for identifying Fusarium damage and the other for identifying DON content. The Fusarium damage classifier had an accuracy of 85% and the DON content classifier had an accuracy of 80%. While a single sample image classification will not replace manual testing, the use of multiple samples from one harvest could reduce manual inspections.

Dissertation (Ph.D.)--Auburn University, / Abstract. Vita. Includes bibliographic references (p.102-108).

Hyperspectral imaging is the collection, processing and analysis of spectral data in numerous contiguous wavelength bands while also providing spatial context. Some of the commonly used instruments for hyperspectral imaging are pushbroom scanning imaging systems, grating based imaging spectrometers and more recently electronically tunable filters. Electronically tunable filters offer the advantages of compactness and absence of mechanically movable parts. Electronically tunable filters have the ability to rapidly switch between wavelengths and provide spatial and spectral information over a large wavelength range. They involve the use of materials whose response to light can be altered in the presence of an external stimulus. While these filters offer some unique advantages, they also present some equally unique challenges. This research work involves the design and development of a multichannel imaging system using electronically tunable Liquid Crystal Fabry-Perot etalons. This instrument is called the Liquid Crystal Hyperspectral Imager (LiCHI). LiCHI images four spectral regions simultaneously and presents a trade-off between spatial and spectral domains. This simultaneity of measurements in multiple wavelengths can be exploited for dynamic and ephemeral events. LiCHI was initially designed for multispectral imaging of space plasmas but its versatility was demonstrated by testing in the field for multiple applications including landscape analysis and anomaly detection. The results obtained after testing of this instrument and analysis of the images are promising and demonstrate LiCHI as a good candidate for hyperspectral imaging. The challenges posed by LiCHI for each of these applications have also been explored.

This thesis focuses on developing a nondestructive strategy for measuring the quality of food using hyperspectral imaging. The specific focus is to develop a classification methodology for detecting bruised/unbruised areas in hyperspectral images of fruits such as strawberries through the classification of pixels containing the edible portion of the fruit. A multiband segmentation algorithm is formulated to generate a mask for extracting the edible pixels from each band in a hypercube. A key feature of the segmentation algorithm is that it makes no prior assumptions for selecting the bands involved in the segmentation. Consequently, different bands may be selected for different hypercubes to accommodate the intra-hypercube variations. Gaussian univariate classifiers are implemented to classify the bruised-unbruised pixels in each band and it is shown that many band classifiers yield 100% classification accuracies. Furthermore, it is shown that the bands that contain the most useful discriminatory information for classifying bruised-unbruised pixels can be identified from the classification results. The strategy developed in this study will facilitate the design of fruit sorting systems using NIR cameras with selected bands.

No description available.

Quantification of spectral data is of great interest in many fields of science, since it can provide further insight into other properties of an object. However, traditional cameras are usually made to image the world in a similar fashion as to how we see it, wherefore they are usually not fit to record nor measure further spectral information. To get a better insight into the spectral properties of an object, a hyperspectral camera might be of use, since those can often identify and measure hundreds of different spectral bands. In this study we look at the construction and further development of an existing design of a push broom hyperspectral imaging system, built with optics for a fraction of the cost of commercial ones. With developed software and objects at hand a spectral calibration was performed, showing a possible spectral range of 184(2)-918(11) nm, but the use of the whole spectral range was however not possible due to limitations in the transmissivity of the lenses below 350 nm. A shift of the spectral range towards longer wavelengths is proposed, which would give further insight into the near infrared spectrum without any information losses. It was found that the spectral calibration of the imager was the main limiting factor of the system, since inaccuracies up to ±11 nm were identified, while the resolution has been found to be 1.4 nm in previous studies, proving that better calibrations are of essence. In good operating conditions, the resolution in the angle of view of the imager was found to be 0.55 mdeg. If the measurement conditions are not as good, or if such kind of spatial resolution is not required, a camera with a smaller detector size and larger pixels could be used to lower the cost of the system without a deterioration in image quality, since the uncertainties in the calibrations and measurement conditions were found to be the limiting factor.

Visual sensing in robotics, especially in the context of autonomous vehicles, has advanced quickly and many important contributions have been made in the areas of target classification. Typical to these studies is the use of the Red-Green-Blue (RGB) camera. Separately, in the field of remote sensing, the hyperspectral camera has been used to perform classification tasks on natural and man-made objects from typically aerial or satellite platforms. Hyperspectral data is characterized by a very fine spectral resolution, resulting in a significant increase in the ability to identify materials in the image. This hardware has not been studied in the context of autonomy as the sensors are large, expensive, and have non-trivial image capture times. This work presents three novel contributions: a Labeled Hyperspectral Image Dataset (LHID) of ground-level, outdoor objects based on typical scenes that a vehicle or pedestrian may encounter, an open-source hyperspectral interface software package (HSImage), and a feature selection based sensor design algorithm for object detection sensors (DLSD). These three contributions are novel and useful in the fields of hyperspectral data analysis, visual sensor design, and hyperspectral machine learning. The hyperspectral dataset and hyperspectral interface software were used in the design and testing of the sensor design algorithm. The LHID is shown to be useful for machine learning tasks through experimentation and provides a unique data source for hyperspectral machine learning. HSImage is shown to be useful for manipulating, labeling and interacting with hyperspectral data, and allows wavelength and classification based data retrieval, storage of labeling information and ambient light data. DLSD is shown to be useful for creating wavelength bands for a sensor design that increase the accuracy of classifiers trained on data from the LHID. DLSD shows accuracy near that of the full spectrum hyperspectral data, with a reduction in features on the order of 100 times. It compared favorably to other state-of-the-art wavelength feature selection techniques and exceeded the accuracy of an RGB sensor by 10%. / Ph. D. / To allow for better performance of autonomous vehicles in the complex road environment, identifying different objects in the roadway or near it is very important. Typically, cameras are used to identify objects and there has been much research into this task. However, the type of camera used is an RGB camera, the same used in consumer electronics, and it has a limited ability to identify colors. Instead, it only detects red, green, and blue and combines the results of these three measurements to simulate color. Hyperspectral cameras are specialized hardware that can detect individual colors, without having to simulate them. This study details an algorithm that will design a sensor for autonomous vehicle object identification that leverages the higher amount of information in a hyperspectral camera, but keep the simpler hardware of the RGB camera. This study presents three separate novel contributions: A database of hyperspectral images useful for tasks related to autonomous vehicles, a software tool that allows scientific study of hyperspectral images, and an algorithm that provides a sensor design that is useful for object identification. Experiments using the database show that it is useful for research tasks related to autonomous vehicles. The software tool is shown to be useful to interfacing between image files, algorithms and external software, and the sensor design algorithm is shown to be comparable to other such algorithms in accuracy, but outperforms the other algorithms in the size of the data required to complete the goal.

[ES] El objetivo de esta tesis doctoral es evaluar la técnica de imagen hiperespectral en el rango visible e infrarrojo cercano, en combinación con técnicas quimiométricas para la evaluación de la calidad de la fruta en poscosecha de manera eficaz y sostenible. Con este fin, se presentan diferentes estudios en los que se evalúa la calidad de algunas frutas que por su valor económico, estratégico o social, son de especial importancia en la Comunidad Valenciana como son el caqui 'Rojo Brillante', la granada 'Mollar de Elche', el níspero 'Algerie' o diferentes cultivares de nectarina. En primer lugar se llevó a cabo la monitorización de la calidad poscosecha de nectarinas 'Big Top' y 'Magique' usando imagen hiperespectral en reflectancia y transmitancia. Al mismo tiempo se evaluó la transmitancia para la detección de huesos abiertos. Se llevó a cabo también un estudio para distinguir los cultivares 'Big Top' y "Diamond Ray", los cuales poseen un aspecto muy similar pero sabor diferente. En cuanto al caqui 'Rojo Brillante', la imagen hiperespectral fue estudiada por una parte para monitorear su madurez, y por otra parte para evaluar la astringencia de esta fruta, que debe ser completamente eliminada antes de su comercialización. Las propiedades físico-químicas de la granada 'Mollar de Elche' fueron evaluadas usando imagen de color e hiperespectral durante su madurez usando la información de la fruta intacta y de los arilos. Finalmente, esta técnica se usó para caracterizar e identificar los defectos internos y externos del níspero 'Algerie'. En la predicción de los índices de calidad IQI y RPI usando imagen en reflectancia y transmitancia se obtuvieron valores de R2 alrededor de 0,90 y en la discriminación por firmeza, una precisión entorno al 95 % usando longitudes de onda seleccionadas. En cuanto a la detección de huesos abiertos, el uso de la imagen hiperespectral en transmitancia obtuvo un 93,5 % de clasificación correcta de frutas con hueso normal y 100 % con hueso abierto usando modelos PLS-DA y 7 longitudes de onda. Los resultados obtenidos en la clasificación de los cultivares 'Big Top' y 'Diamond Ray' mostraron una fiabilidad superior al 96,0 % mediante el uso de modelos PLS-DA y 14 longitudes de onda seleccionadas, superando a la imagen de color (56,9 %) y a un panel entrenado (54,5 %). Con respecto al caqui, los resultados obtenidos indicaron que es posible distinguir entre tres estados de madurez con una precisión del 96,0 % usando modelos QDA y se predijo su firmeza obteniendo un valor de R2 de 0,80 usando PLS-R. En cuanto a la astringencia, se llevaron a cabo dos estudios similares en los que en el primero se discriminó la fruta de acuerdo al tiempo de tratamiento con altas concentraciones de CO2 con una precisión entorno al 95,0 % usando QDA. En el segundo se discriminó la fruta de acuerdo a un valor de contenido en taninos (0,04 %) y se determinó qué área de la fruta era mejor para realizar esta discriminación. Así se obtuvo una precisión del 86,9 % usando la zona media y 23 longitudes de onda. Los resultados obtenidos para la granada indicaron que la imagen de color e hiperespectral poseen una precisión similar en la predicción de las propiedades fisicoquímicas usando PLS-R y la información de la fruta intacta. Sin embargo, cuando se usó la información de los arilos, la imagen hiperespectral fue más precisa. En cuanto a la discriminación del estado de madurez usando PLS-DA, la imagen hiperespectral ofreció mayor precisión, 95,0 %, usando la información de la fruta intacta y del 100 % usando la de los arilos. Finalmente, los resultados obtenidos para el níspero indicaron que la imagen hiperespectral junto con el método de clasificación XGBOOST pudo discriminar entre muestras con y sin defectos con una precisión del 97,5 % y entre muestras sin defectos o con defectos internos o externos con una precisión del 96,7 %. Además fue posible distinguir entre los dife / [CA] L'objectiu de la present tesi doctoral se centra en avaluar la capacitat de la imatge hiperespectral en el rang visible i infraroig pròxim, en combinació amb mètodes quimiomètrics, per a l'avaluació de la qualitat de la fruita en post collita de manera eficaç i sostenible. A aquest efecte, es presenten diferents estudis en els quals s'avalua la qualitat d'algunes fruites que pel seu valor econòmic, estratègic o social, són d'especial importància a la Comunitat Valenciana com són el caqui 'Rojo Brillante', la magrana 'Mollar de Elche', el nispro 'Algerie' o diferents cultivares de nectarina. En primer lloc es va dur a terme la monitorització de la qualitat post collita de nectarines 'Big Top' i 'Magique' per mitjà d'imatge hiperespectral en reflectància i trasnmitancia. Així mateix es va avaluar la transmitància per a la detecció d'ossos oberts. Es va dur a terme també un estudi per distingir els cultivares 'Big Top' i 'Diamond Ray', els quals posseeixen un aspecte molt semblant però sabor diferent. Pel que fa al caqui 'Rojo Brillante', la imatge hiperespectral va ser estudiada d'una banda per a monitoritzar la seua maduresa, i per un altre costat per avaluar l'astringència, que ha de ser completament eliminada abans de la seua comercialització. Les propietats fisicoquímiques de la magrana 'Mollar de Elche' van ser avaluades per la imatge de color i hiperespectral durant la seua maduresa usant la informació de la fruita intacta i els arils. Finalment, aquesta tècnica es va fer servir per caracteritzar i identificar els defectes interns i externs del nispro 'Algerie'. En la predicció dels índexs de qualitat IQI i RPI usant imatge en reflectància com en trasnmitancia es van obtindre valors de R2 al voltant de 0,90 i en la discriminació per fermesa una precisió entorn del 95,0 % utilitzant longituds d'ona seleccionades. Pel que fa a la detecció d'ossos oberts, l'ús de la imatge hiperespectral en transmitància va obtindre un 93,5 % classificació correcta de fruites amb os normal i 100 % amb os obert usant models PLS-DA i 7 longituds d'ona. Els resultats obtinguts en la classificació dels cultivares 'Big Top' i 'Diamond Ray' van mostrar una fiabilitat superior al 96,0 % per mitjà de l'ús de models PLS-DA i 14 longituds d'ona, superant a la imatge de color (56,9 %) i a un panell sensorial entrenat (54,5 %). Quant al caqui, els resultats obtinguts van indicar que és possible distingir entre tres estats de maduresa amb una precisió del 96,0 % usant models QDA i es va predir la seua fermesa obtenint un valor de R2 de 0,80 usant PLS-R. Pel que fa a l'astringència, es van dur a terme dos estudis similars en què el primer es va discriminar la fruita d'acord al temps de tractament amb altes concentracions de CO2 amb una precisió al voltant del 95,0 % usant QDA. En el segon, es va discriminar la fruita d'acord a un valor de contingut en tanins (0,04 %) i es va determinar quina part de la fruita era millor per a realitzar aquesta discriminació. Així es va obtindre una precisió del 86,9 % usant la zona mitjana i 23 longituds d'ona. Els resultats obtinguts per la magrana van indicar que la imatge de color i hiperespectral posseïxen una precisió semblant a la predicció de les propietats fisicoquímiques usant PLS-R i la informació de la fruita intacta. No obstant això, quan es va usar la informació dels arils, la imatge hiperespectral va ser més precisa. Quant a la discriminació de l'estat de maduresa usant PLS-DA, la imatge hiperespectral va oferir major precisió (95,0 %) usant la informació de la fruita intacta i del 100 % usant la dels arils. Finalment, els resultats obtinguts pel nispro indiquen que la imatge hiperespectral juntament amb el mètode de classificació XGBOOST va poder discriminar entre mostres amb i sense defectes amb una precisió del 97,5 % i entre mostres sense defectes o amb defectes interns o externs amb una precisió del 96,7 %. A més, va ser possible distingir entre / [EN] The objective of this doctoral thesis is to evaluate the potential of the hyperspectral imaging in the visible and near infrared range in combination with chemometrics for the assessment of the postharvest quality of fruit in a non-destructive, efficient and sustainable manner. To this end, different studies are presented in which the quality of some fruits is evaluated. Due to their economic, strategic or social value, the selected fruits are of special importance in the Valencian Community, such as Persimmon 'Rojo Brillante', the pomegranate 'Mollar de Elche', the loquat 'Algerie' or different nectarine cultivars. First, the quality monitoring of 'Big Top' and 'Magique' nectarines was carried out using reflectance and transmittance images. At the same time, transmittance was evaluated for the detection of split pit. In addition, a classification was performed to distinguish the 'Big Top' and 'Diamond Ray' cultivars, which look very similar but have different flavour. Whereas that for the 'Rojo Brillante' persimmon, the hyperspectral imaging was studied on the one hand to monitor its maturity, and on the other hand to evaluate the astringency of this fruit, which must be completely eliminated before its commercialization. The physicochemical properties of the 'Mollar de Elche' pomegranate were evaluated by means of hyperspectral and colour imaging during its maturity using the information from the intact fruit and arils. Finally, this technique was used to characterise and identify the internal and external defects of the 'Algerie' loquat. In the prediction of the IQI and RPI quality indexes using reflectance and transmittance images, R2 values around 0.90 were obtained and in the discrimination according to firmness, accuracy around 95.0 % using selected wavelengths was obtained. Regarding the split pit detection, the use of the hyperspectral image in transmittance mode obtained a 93.5 % of fruits with normal bone correctly classified and 100% with split pit using PLS-DA models and 7 wavelengths. The results obtained in the classification of 'Big Top' and 'Diamond Ray' fruits show accuracy higher than 96.0 % by using PLS-DA models and 14 selected wavelengths, higher than the obtained with colour image (56.9 %) and a trained panel (54.5 %). According to persimmon, the results obtained indicated that it is possible to distinguish between three states of maturity with an accuracy of 96.0 % using QDA models and its firmness was predicted obtaining a R2 value of 0.80 using PLS-R. Regarding astringency, two similar studies were carried out. In the first study, the fruit was classified according to the time of treatment with high concentrations of CO2 with a precision of around 95.0 % using QDA. In the second, the fruit was discriminated according to a threshold value of soluble tannins (0.04 %) and was determined what fruit area was better to perform this discrimination. Thus, an accuracy of 86.9 % was obtained using the middle area and 23 wavelengths. The results obtained for the pomegranate indicated that the use of colour and hyperspectral images have a similar precision in the prediction of physicochemical properties using PLS-R and the intact fruit information. However, when the information from the arils was used, the hyperspectral image was more accurate. Regarding the discrimination by the state of maturity using PLS-DA, the hyperspectral image offered greater precision, of 95.0 % using the information from the intact fruit and 100 % using that from the arils. Finally, the results obtained for the 'Algerie' loquat indicated that the hyperspectral image with the XGBOOST classification method could discriminate between sound samples and samples with defects with accuracy of 97.5 % and between sound samples or samples with internal or external defects with an accuracy of 96.7 %. It was also possible to distinguish between the different defects with an accuracy of 95.9 %. / Munera Picazo, SM. (2019). Application of hyperspectral imaging combined with chemometrics for the non-destructive evaluation of the quality of fruit in postharvest [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/125954 / TESIS

No description available.