Browsing by Author "Nturambirwe, Jean Frederic Isingizwe"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Classification learning of latent bruise damage to apples using shortwave infrared hyperspectral imaging
    (MDPI, 2021-07-22) Nturambirwe, Jean Frederic Isingizwe; Perold, Willem Jacobus; Opara, Umezuruike Linus
    Bruise damage is a very commonly occurring defect in apple fruit which facilitates disease occurrence and spread, leads to fruit deterioration and can greatly contribute to postharvest loss. The detection of bruises at their earliest stage of development can be advantageous for screening purposes. An experiment to induce soft bruises in Golden Delicious apples was conducted by applying impact energy at different levels, which allowed to investigate the detectability of bruises at their latent stage. The existence of bruises that were rather invisible to the naked eye and to a digital camera was proven by reconstruction of hyperspectral images of bruised apples, based on effective wavelengths and data dimensionality reduced hyperspectrograms. Machine learning classifiers, namely ensemble subspace discriminant (ESD), k-nearest neighbors (KNN), support vector machine (SVM) and linear discriminant analysis (LDA) were used to build models for detecting bruises at their latent stage, to study the influence of time after bruise occurrence on detection performance and to model quantitative aspects of bruises (severity), spanning from latent to visible bruises. Over all classifiers, detection models had a higher performance than quantitative ones. Given its highest speed in prediction and high classification performance, SVM was rated most recommendable for detection tasks. However, ESD models had the highest classification accuracy in quantitative (>85%) models and were found to be relatively better suited for such a multiple category classification problem than the rest.
  • Thumbnail Image
    Item
    Non-destructive measurement of internal fruit quality using SQUID-NMR techniques
    (Stellenbosch : Stellenbosch University, 2012-12) Nturambirwe, Jean Frederic Isingizwe; Perold, W. J.; Opara, L.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.; Stellenbosch University. Faculty of Agrisciences. Dept.of Horticulture
    ENGLISH ABSTRACT: The SQUID-NMR technique has been increasingly recommended by many researchers as holding a lot of potential, and it is believed it will become an invaluable tool for non-destructive evaluation in the future. Most of its potential is yet to be exploited. Non-destructive quality control of food products is one of the applications where such a system is being tried. Much of the progress that has been made in improving such a system to the present degree of user friendliness and cost effectiveness shows that, with more effort, it would be possible to implement the technology for on-line sorting, and possibly to reduce it down to hand-held devices. The goal was to investigate the feasibility of the internal fruit quality measurements using the NMR technique, and to develop a SQUID system suitable for SQUID-NMR application, intended for a later integration in a full SQUID-NMR system. A working dc SQUID was manufactured on an YBCO (Y Ba2Cu3O7􀀀 ) thin film deposited on a 10 mm x 10 mm MgO substrate. The SQUID was made of microbridge Josephson junctions, patterned by using the double resist laser lithography method, implemented during the course of this manufacturing process. The test of the SQUID showed a non-hysteretic current-voltage characteristic. Under the action of bringing a magnet closer to the SQUID under test, and then retracting it, the modulation of the I-V curve was observed. The critical current of the SQUID was 20 A and the resistance was 5.5 A series of experiments were performed on destructive measurements of the sugar content in table grapes using NMR, in order to evaluate the feasibility of this technique. The total sugars(TSS) measurements of the same samples were carried out by refractometry, chosen as a conventional method for validation. The NMR measurements were evaluated to be 5.4% precise and have an accuracy of 9.3% relative to the refractometry measurements. A further series of experiments were carried out on a high-Tc SQUID-NMR system. A high correlation coefficient (0.85) of the increasing values of the T1 and T2 relaxation times to the decreasing concentration of sugar (sucrose) in water was obtained. Non-destructive measurements T1 and T2 in table grapes suggested a possible prediction of sugar content in table grapes from the values of T1 or T2. This technique also presented many advantages compared to the conventional high field NMR technique, such as the fast measurements that do not require spectral processing, the ease of sample preparation, and its non-destructive nature.
  • Thumbnail Image
    Item
    Non-destructive measurement of internal quality of apple fruit by a contactless NIR spectrometer with genetic algorithm model optimization
    (Elsevier, 2019) Nturambirwe, Jean Frederic Isingizwe; Nieuwoudt, Helene H.; Perold, Willem J.; Opara, Umezuruike Linus
    ENGLISH ABSTRACT: Spectrometric methods based on near infrared radiation (NIR) are commonly used effectively in the agricultural and food industry. However, these methods still face limitations whereby meeting requirements for application such as nondestructive quality testing of large fruits and automated sorting and grading is still a challenge. A Fourier transform (FT)-NIR spectrometer (emission head, EH mode of Matrix-F) that simulates on-line sample scanning (contactless, large sample size (100 mm)) was used to predict internal properties of apple fruit. The EH was compared to laboratory multipurpose analyzer (MPA) FT-NIR spectrometer using two contact-sample presentation modes with relatively smaller sample size (≤22 mm); namely, the integrating sphere (IS) and the solid probe (SP). Three apple cultivars (Golden Delicious, Granny Smith and Royal Gala) sourced from two retail stores (in Stellenbosch, South Africa) were used to constitute variability in the sample set. Partial least squares regression (PLSR) prediction models for internal quality (total soluble solids (TSS) and titratable acidity (TA)) were developed and validated on external test samples in various scenarios. Genetic algorithm (GA) based optimization of PLS models was used to produce optimal models prior to instrumental comparison. Model optimization using GA improved performance by a margin of 30% of the original root mean square error of cross validation for the contactless system bringing it closer to the performance of models from the MPA. The EH's performance makes it an attractive option for achieving on-line application of NIR spectroscopy for sorting apples based on internal quality.