Daniel E. Guyer

Use of genetic artificial neural networks and spectral imaging for defect detection on cherries

Abstract A machine vision system was created to identify different types of tissue characteristics on cherries. It consists of an enhanced NIR range vidicon black and white camera (sensing range 400–2000 nm), a monochrometer controlled light source, and a computer. Multiple spectral images of cherry samples were collected over the 680–1280 nm range at increments of 40 nm. Using the spectral signatures of different tissues on cherry images, artificial neural networks were applied to pixel-wise classification. An enhanced genetic algorithm was applied to design the topology and evolve the weights for multi-layer feed forward artificial neural networks. An average of 73% classification accuracy was achieved for correct identification as well as quantification of all types of cherry defects. No false positives or false negatives occurred, errors resulted only from misclassification of defect type or quantification of defect.

APPLE SORTING USING ARTIFICIAL NEURAL NETWORKS AND SPECTRAL IMAGING

Empire and Golden Delicious apples were sorted based on their surface quality conditions using backpropagation neural networks. Pixel gray values and texture features obtained from the entire apple image were used as input to artificial neural network classifiers. Two classification applications were performed: a 2–class classification that included a defective (or stem/calyx) apple group and a good apple group, and a 5–class classification that included all the defective and good apple groups. Effective image resolution was evaluated to shorten the training and testing times in classification with neural networks. Resolution size of 60 U 80 pixels was identified to be efficient and used in all of the classification applications. Effective spectral bands for identification of specific surface characteristics were determined in the 2–class and 5–class classification applications. Artificial neural network classifiers successfully separated apples with defects from non–defective apples without confusing the stem/calyx with defects. Classification success in the 2–class classification ranged from 89.2% to 100%. In the 5–class classification, classification success for Empire apples was between 93.8% and 100%, while classification success for Golden Delicious apples was between 89.7% and 94.9% based on the features used.

Pyruvate and flavor development in macerated onions (Allium cepa L.) by γ-glutamyl transpeptidase and exogenous C-S lyase

Abstract Effect of γ-glutamyl transpeptidase in conjunction with exogenous C-S lyase on pyruvate content in macerated onion and flavor profile was studied. Pyruvate production of 2.5-fold greater than that of the control was obtained in γ-glutamyl transpeptidase and exogenous C-S lyase treated onions held for 20 h at 37 °C. Relative abundance of flavor compounds in Spartan Banner, a pungent onion, varied from a yellow sweet salad-type onion. The effect of γ-glutamyl transpeptidase and exogenous C-S lyase on onion flavor profile was shown by a shift of major components from methyl propyl disulfide, methyl propenyl trisulfide, dimethyl tetrasulfide and propyl 1-propenyl trisulfide, into new major components, methyl 1-propenyl disulfide, dipropyl disulfide, propyl 1-propenyl disulfide, methyl 1-propenyl trisulfide, and propyl 1-propenyl trisulfide. This increase in 1-propenyl containing flavor compounds may effect overall flavor of γ-glutamyl transpeptidase and exogenous C-S lyase treated onion extracts.

Rapid and/or nondestructive quality evaluation methods for potatoes

Several non-invasive techniques, studied for quality evaluation of potatoes, are reviewed.The application of non-destructive methods on potato sorting are discussed in detail.Quality estimation studies of French fries and chips via electronic measurements are discussed.Discussion of the state of art commercially available systems potato sorting systems is provided.Possibility of applying emerging technologies in potatoes postharvest and industry is studied. Potato, with its several processed products, has a major rank on the human diet in many countries. Among electronic-based methods that have been used for tracking and rapidly measuring the quality attributes of raw and processed potatoes (more specifically: French fry and chip), and which are reviewed herein, vision and spectroscopic systems have shown the most promising applicability, robustness, and stable performance. Detection of external and internal defects associated with potatoes during harvesting and handling operations has been made possible using nondestructive techniques. Commercial electronic systems used for sorting potato tubers and products are being incorporated into the potato industry and are included and discussed in detail related to operation theories and performance. The need for healthy food requires more attention for detecting harmful chemicals in fried products such as acrylamide which itself demands continuous tracking of sugar during storage and in French fries or chips which encourages the need for constitute-based sorting for potatoes. Hyperspectral imaging is one of the most recently emerging tools and provides advantages of vision and spectroscopic systems and can be used, after speeding up image acquisition time, in prediction of processing-related constituents as well as defects detection. Moreover other noninvasive techniques, such as NMR and X-ray CT, have shown the potential for successful application in quality monitoring of fruits and vegetables with expected possibility for application in potatoes.

Tissue reflectance and machine vision for automated sweet cherry sorting

This study describes machine vision procedures which are able to classify defective cherries from non-defective cherries. Defects can be divided into bruises, dry cracks, and wet cracks. Bandpass filters that enhance the intensity contrast between bruised and unbruised cherries are determined. An optimum combination of two wavelengths is identified at 750 nm (near-infrared range) and 500 nm (green range). An optimum single wavelength is identified at 750 nm. The image acquisition using these filters is described. Four detection methods using single view infrared images were studied. One method performed well in classifying cherries with bruises and wet cracks from non-defective cherries. One detection method using single view green images is studied. It performs well in classifying cherries with dry cracks from non-defective cherries. One detection method using infrared images and another using green images are used in combination to perform the detection on the entire surface of cherries. Two images, infrared and green, are taken from each of 6 orthogonal directions from the cherries. The integrated classifier misclassified 13% of non-defective cherries, 16% of bruised cherries, 0% of cherries with wet cracks, and 10% of cherries with dry cracks.

Effects of supercritical carbon dioxide conditions on onion oil desorption

Abstract Desorption of onion oil from adsorbent beds loaded with one liter of onion juice by supercritical carbon dioxide at different pressure, temperature, and density was investigated. At constant temperature; when density was increased from 0.69 to 0.86 g/ml at 37 °C, the gravimetric yield of onion oil by percentage was increased from 0.0112% to 0.0222%. When density was increased from 0.79 to 0.86 g/ml at 50 °C, the gravimetric yield of onion oil by percentage was increased from 0.0189% to 0.0231%. At constant density of 0.86 g/ml, an increase in temperature from 37 to 50 °C increased the gravimetric yield of onion oil by percentage from 0.0222% to 0.0231%. Yield difference was significant ( p =0.05) only for the 0.69–0.86 g/ml density difference. Similar composition onion oil desorbed at the different supercritical carbon dioxide conditions was found by using GC-MS in the analysis of onion oil.

Automated insect classification with combined global and local features for orchard management

The paper describes an image-based system to identify certain insects in orchards using image processing and pattern recognition methods. The insect samples for analysis are eight common species in Michigan orchards. Global features, including geometric, contour, moment, texture, and color features, were calculated under gray level, boundary, and binary transformations of the original image. Global features are sensitive to clutter, overlapping, occlusion and other partial information loss situations while local features are more stable under these situations. For local features, a scale invariant transform feature (SIFT) was used to describe the neighborhood of each interest point detected by the Difference of Gaussian function. A bag of words representation method was introduced to convert SIFT vectors to be normalized histogram vectors for training classifiers. A hierarchical combination model was established to combine the global features-based classification and local features-based classification by means of cascading. Results of original images and partial information loss insect images classification showed the combination of global and local features classification can achieve accuracy of 96.9% on original images and improve partial information loss image classification by 4% when comparing against only using global features.

Integrating reflectance and fluorescence imaging for apple disorder classification

Multispectral imaging in reflectance and fluorescence modes combined with neural network analysis was used to classify various types of apple disorder from three apple varieties (Honey Crisp, Red Cort, and Red Delicious). Eighteen images from a combination of filter sets and three different imaging modes (reflectance, visible light induced fluorescence, and UV induced fluorescence) were acquired for each apple sample as a base for pixel-level classification into normal or disorder tissue. Two classification schemes, a 2-class and a multiple class, were developed and tested in this study. In the 2-class scheme, pixels were categorized into normal or disorder tissue, whereas in the multiple class scheme, pixels were categorized into normal, bitter pit, black rot, decay, soft scald, and superficial scald tissues. Results indicate that single variety training under the 2-class scheme yielded highest accuracy with total accuracy of 95, 97, and 100 % for Honey Crisp, Red Cort, or Red Delicious respectively. In the multiple-class scheme, the classification accuracy of Honey Crisp apple for normal, bitter pit, black rot, decay, and soft scald tissue was 94, 93, 97, 97, and 94 % respectively. Through variable selection analysis, in the 2-class scheme, fluorescence models yielded higher total classification accuracy compared to reflection models. For Red Cort and Red Delicious, models with only FUV yield more than 95% classification accuracy, demonstrating a potential of fluorescence to detect superficial scald. Several important wavelengths, including 680, 740, 905 and 940 nm, were identified from the filter combination analysis. The results indicate the potential of this technique to accurately recognize different types of disorder on apple.

Classification of processing asparagus sections using color images

Computer vision and pattern recognition tool to determine asparagus sections.Essential information to sort asparagus based on their section is offered.Method that could be applied in an in-line quality sorter of asparagus. Impartial classification of Asparagus sections (Asparagus officinalis L.), for the purpose of obtaining desired tip to stem pieces ratio in final product, is extremely important to the processing industry. Thus, there is a need to develop a technique that is able to objectively discern between tip and stem pieces, after asparagus has been processed (cut). In this article, a computer vision methodology is proposed to sort asparagus into three classes: tips, mid-stem pieces and bottom-stem pieces. Nine hundred and fifty-five color images from 50mm length asparagus pieces (cuts) for the three different classes were acquired, using a flat panel scanner. After preprocessing, a total of 1931 color, textural, and geometric features were extracted from each color image. The most relevant features were selected using a sequential forward selection algorithm. Forty-three features were found to be effective in designing a neural-network classifier with a 4-fold cross-validated overall performance accuracy of 90.2% (ź2.2%). Results showed that this method is an accurate, reliable, and objective tool to discern between asparagus tips, mid-stem and bottom pieces, and might be applicable to in-line sorting systems.

Integrating Reflectance and Fluorescence Imaging for Apple Disorder Classification

Multispectral imaging in reflectance and fluorescence modes combined with neural network analysis was used to classify various types of apple disorder from three apple varieties (Honey Crisp, Red Cort, and Red Delicious). Eighteen images from a combination of filter sets and three different imaging modes (reflectance, visible light induced fluorescence, and UV induced fluorescence) were acquired for each apple sample as a base for pixel classification into normal or disorder tissue. Two classification models, a 2-class model and a 6-class model, were developed and tested in this study. In the 2-class model, pixels were categorized into normal or disorder tissue, whereas in the 6-class model, pixels were categorized into normal, bitter pit, black rot, decay, soft scald, and superficial scald tissues. Results indicate that single variety training under the 2-class model yielded highest accuracy with total accuracy of 95, 97, and 100 % for Honey Crisp, Red Cort, or Red Delicious respectively. In the 6-class model, the classification accuracy of Honey Crisp apple for normal, bitter pit, black rot, decay, and soft scald tissue was 94, 93, 97, 97, and 94 % respectively. The results indicate the potential of this technique to accurately recognize different types of disorder on apple.