Stephen J. Symons

COMPARISON OF STATISTICAL AND NEURAL NETWORK METHODS FOR CLASSIFYING CEREAL GRAINS USING MACHINE VISION

Two statistical and one neural network classifiers were applied and empirically compared for the classification cereal grain kernels (e.g., Canadian Western Red Spring (CWRS) wheat, Canadian Western Amber Durum (CWAD) wheat, barley, rye, and oats) and for the classification of healthy and six types of damaged (e.g. broken, grass-green/green-frosted, black-point/smudge, mildewed, heated, and bin/fire-burnt) CWRS wheat kernels, using selected morphological and color features extracted from the grain sample images. For the classification of cereal grain kernels and the classification of healthy and damaged CWRS wheat kernels, the k-nearest neighbor statistical classifier and the multilayer neural network (MNN) classifier gave similar and the best classification results. Using a k-nearest neighbor classifier with a selected set of 15 morphological and 13 color features, the average classification accuracies were 98.2, 96.9, 99.0, 98.2, and 99.0% for CWRS wheat, CWAD wheat, barley, rye, and oats, respectively, when trained and tested with three different training and testing data sets. Using a k-nearest neighbor classifier with a selected set of 24 color and four morphological features, the average classification accuracies were 92.5 (healthy), 90.3 (broken), 98.6 (mildewed), 99.0 (grass-green/green-frosted), 99.1 (black-point/smudged), 97.5 (heated), and 100.0% (bin/fire-burnt), respectively, when trained and tested with three different training and testing data sets. The classification accuracies achieved using a parametric classifier were lower than the classification accuracies achieved using both the k-nearest neighbor and the MNN classifiers.

Developments in the use of image analysis for the assessment of oriental noodle appearance and colour

Canada Prairie Spring Red (8 cultivars) and Canada Prairie Spring White (10 cultivars) wheat were milled on an Allis Chalmers mill to yield straight grade flours of a constant 74% extraction rate. These flours were used to prepare both fresh yellow alkaline and white salted noodles. The appearance of the noodles prepared from the white and red seed coated wheats was measured using two scanner based image analysis systems differing in their means of colour correction. Significant differences in the number of detectable specks (p<0.0001) within both alkaline and white salted noodles were observed between noodles prepared from either white and red seed coated wheat flour using either system. Speck size and gray level density parameter combinations were found to influence the red, green and blue, RGB, values of the detected specks. Unique differences in both alkaline and white salted noodle colour components were observed using the inexpensive, colour calibrated colour system for noodles prepared from either white or red seed coated wheat flour. These differences were detectable at both 2 and 24 h after production offering noodle manufacturer’s an unique, inexpensive means of monitoring their noodle’s appearance.

Color calibration of scanners for scanner-independent grain grading

ABSTRACT Scanner technology is emerging as a cost-effective and robust imaging alternative to camera-based systems in many applications. However, scanner technology is changing so fast that image quality can vary from model to model. It is critical that images scanned with different scanners be brought to a common basis for processing and measurement through a calibration process that eliminates scanner-to-scanner variability. The focus of this research was to investigate scanner-to-scanner variability and develop color correction or mapping functions to allow for machineindependent grain inspection. Various makes and models of scanners were compared for optical and color characteristics. Three different color correction methods wereevaluated: grayscale (GS) transformation, redgreen-blue (RGB) transformation, and histogram matching. All three models of color correction worked within satisfactory tolerance for a multicolor Q60 chart. However, for grain samples of a limited color range, the histogram matchi...

Measurement of Hard Vitreous Kernels in Durum Wheat by Machine Vision

ABSTRACT An imaging method that detects nonvitreous regions in sound kernels of durum wheat at high speed is described. Kernels are analyzed simultaneously for individual vitreousness and individual kernel size and shape are measured concurrently. The measurement of 500 kernels per sample is adequate for highly reproducible results. Significant agreement was found between inspector-determined hard vitreous kernel percentages (HVK) and machine-determined HVK scores for export cargo samples of Canadian Western Amber Durum (CWAD), with differences between the two methods of typically ±3%. For railcar samples of CWAD taken on delivery to the terminal, agreement between inspector-determined and machine-determined HVK scores were more variable. The variability between the two methods generally increased as the HVK score of the sample became lower. For inspector-determined HVK scores of <50%, difference between inspector and machine HVK scores for some samples was substantial. Such large differences are partiall...

Influence of Sprout Damage on Oriental Noodle Appearance as Assessed by Image Analysis

ABSTRACT Fresh alkaline (kansui) and white salted noodles prepared from sound and sprout damaged patent flours of the western Canadian wheat class Canadian Prairie Spring White (CPSW) cv. Vista were characterized by image analysis (IA). In all samples, the number of discolored spots increased with aging <24 hr (24 ± 1°C), although the number of spots per sample was significantly influenced by the degree of sprout damage. Alkaline kansui noodles made from severely sprouted wheat (Day 5) flours had the greatest number of spots per image at 1 hr (114) and increased to 256 spots per image by 7 hr. This represented an approximate fivefold greater number of spots as compared with the sound flour kansui noodle at 7hr. No further increase in spot numbers was detected in the severely sprouted sample with aging for 24 hr. Significantly fewer spots were observed in the white salted noodles (WSN) prepared from heavily sprouted wheat with 29 spots per image at 1 hr increasing to only 54.5 after 24 hr. The IA system wa...

Image Analysis of Asian Noodle Appearance: Impact of Hexaploid Wheat with a Red Seed Coat

ABSTRACT Fresh alkaline (kansui) and white salted noodles (WSN) prepared from patent and straight-grade flour of the western Canadian wheat class Canadian Red Spring Wheat (CWRS) were visually characterizedby image analysis (IA) over a 24-hr period. In both kansui noodles, the number of spots increased over time, while minimal change was detected in the WSN prepared from either flour. Maximum spot generation was observed in the straight-grade kansui noodles, increasing from 53.1 spots per image at 1 hr to 76.2 by 7 hr before declining to 55.9 at 24 hr. Significant differences were detected in the number of detectable spots among the kansui, patent, and straight-grade noodles over the initial 7 hr, but by 24 hr no discernible differences were observed. Fewest noodle spots were observed in the patent WSN (10.1 spots per image at 1 hr rising to 13.3 at 24 hr). Hence, straight-grade flours yielded more spots than the matching patent flours, while WSN had consistently fewer spots relative to the kansui noodles...

Detection of Hard Vitreous and Starchy Kernels in Amber Durum Wheat Samples Using Hyperspectral Imaging (GRL Number M306)

The Grain Research laboratory (GRl) of the Canadian Grain Com mission (CGC) is committed to developing objective instrumen tal methods for grading grains and quality assessment of grain products. historically, method development at the GRl has been carried out through independent research in the technological areas of conventional image analysis and near infrared (NIR) spec troscopy. NIR spectroscopy has been suc cessfully used to determine moisture and protein contents in grains 1,2 while imaging

Lentil seed size distribution with machine vision

In this study, a machine vision system was developed to determine size distribution of lentil seeds from images of bulk samples. Seed boundaries were separated using morphological processing techniques. The size measurements were partitioned into various size categories in terms of histograms. The system was evaluated on different lentil varieties that differed in both seed size and colour characteristics. The vision results were compared with the caliper and sieve measurements to determine system performance. Results of this study indicate that machine vision can be used to obtain seed size distribution from images of bulk lentils.

c-Myc-Induced Extrachromosomal Elements Carry Active Chromatin

Murine Pre-B lymphocytes with experimentally activated MycER show both chromosomal and extrachromosomal gene amplification. In this report, we have elucidated the size, structure, and functional components of c-Myc-induced extrachromosomal elements (EEs). Scanning electron microscopy revealed that EEs isolated from MycER-activated Pre-B+ cells are an average of 10 times larger than EEs isolated from non-MycER-activated control Pre-B- cells. We demonstrate that these large c-Myc-induced EEs are associated with histone proteins, whereas EEs of non-MycER-activated Pre B- cells are not. Immunohistochemistry and Western blot analyses using pan-histone-specific, histone H3 phosphorylation-specific, and histone H4 acetylation-specific antibodies indicate that a significant proportion of EEs analyzed from MycER-activated cells harbors transcriptionally competent and/or active chromatin. Moreover, these large, c-Myc-induced EEs carry genes. Whereas the total genetic make-up of these c-Myc-induced EEs is unknown, we found that 30.2% of them contain the dihydrofolate reductase (DHFR) gene, whereas cyclin C (CCNC) was absent. In addition, 50% of these c-Myc-activated Pre-B+ EEs incorporated bromodeoxyuridine (BrdU), identifying them as genetic structures that self-propagate. In contrast, EEs isolated from non-Myc-activated cells neither carry the DHFR gene nor incorporate BrdU, suggesting that c-Myc deregulation generates a new class of EEs.

Seed Sizing with Image Analysis

Seed size is an important grading factor in pulse grains. Seed processors and canning industry prefer shipments with minimal size variability to those with a wider range of seed distribution. Typically, seed size distribution of a batch of grains is determined by sieving a representative sample from a batch. Image analysis based seed sizing provides a faster, more consistent, accurate and effortless alternative to sieving. The objective of this study was to develop a machine vision system for sizing seeds of various shapes. A flatbed scanner based image analysis application was developed to size circular (peas), elliptical (soybean) and multifaceted (chickpeas) shaped seeds by imaging a bulk poured sample. This application automatically separates the seed boundaries in an image, measures individual seeds, and reports size distribution for user-selectable sieve combination in metric or imperial units. Development of the image analysis system along with its performance in comparison with manual sieving is discussed in this paper.