Suranjan Panigrahi

Odorant binding protein based biomimetic sensors for detection of alcohols associated with Salmonella contamination in packaged beef.

Detection of food-borne bacteria present in the food products is critical to prevent the spread of infectious diseases. Intelligent quality sensors are being developed for detecting bacterial pathogens such as Salmonella in beef. One of our research thrusts was to develop novel sensing materials sensitive to specific indicator alcohols at low concentrations. Present work focuses on developing olfactory sensors mimicking insect odorant binding protein to detect alcohols in low concentrations at room temperature. A quartz crystal microbalance (QCM) based sensor in conjunction with synthetic peptide was developed to detect volatile organic compounds indicative to Salmonella contamination in packaged beef. The peptide sequence used as sensing materials was derived from the amino acids sequence of Drosophila odorant binding protein, LUSH. The sensors were used to detect alcohols: 3-methyl-1-butanol and 1-hexanol. The sensors were sensitive to alcohols with estimated lower detection limits of <5 ppm. Thus, the LUSH-derived QCM sensors exhibited potential to detect alcohols at low ppm concentrations.

A GENERALIZED REGRESSION NEURAL NETWORK AND ITS APPLICATION FOR LEAF WETNESS PREDICTION TO FORECAST PLANT DISEASE

Abstract The objective of this paper was the development of an optimal generalized regression neural network (GRNN) for leaf wetness prediction. The GRNN prediction results were compared to those obtained with the standard multiple linear regression (MLR). Leaf wetness, which is difficult to measure directly, has an important effect on the development of disease on plants. In this study, leaf wetness was predicted from micrometeorological factors (temperature, relative humidity, wind speed, solar radiation and precipitation). Simulations showed than the MLR provided an average absolute prediction error of 0.1300 for the training set and 0.1414 for the test set. The GRNN provided an average absolute prediction errors of 0.0491 and 0.0894 on the same data sets, respectively. This error is very low since the leaf wetness initially varies between 0 and 1. The optimized GRNN, therefore, outperformed the MLR in terms of the prediction accuracy. However, the GRNN required more computational time than the MLR. The main disadvantage of the MLR is that it assumes a linear relationship between the feature to be predicted and the measured features. The GRNN automatically extracts the appropriate regression model (linear or nonlinear) from the data.

SPOILAGE IDENTIFICATION OF BEEF USING AN ELECTRONIC NOSE SYSTEM

A commercially available Cyranose-320. conducting polymer-based electronic nose system was used to analyze the volatile organic compounds emanating from fresh beef strip loins (M. Longisimmus lumborum) stored at 4°C and 10°C. Two statistical techniques, i.e., linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA), were used to develop classification models from the collected sensor signals. The performances of the developed models were validated by two different methods: leave-1-out cross-validation, and bootstrapping. The developed models classified meat samples based on the microbial population into “unspoiled” (microbial counts 6.0 log10 cfu/g). Overall, quadratic discriminant-based classification models performed better than linear discriminant analysis based models. For the meat samples stored at 10°C, the highest classification accuracies obtained by the LDA method with leave-1-out and bootstrapping validations were 87.10% and 85.87%, respectively. On the other hand, classification by QDA and subsequent validation by leave-1-out and bootstrapping provided highest accuracies of 87.5% and 97.38%, respectively. For samples stored at 4°C, the LDA method provided highest classification accuracies of 79.17% and 85.64% using leave-1-out and bootstrapping validation, respectively. When the QDA method was used, the highest classification accuracies obtained for the samples stored at 4°C were 87.50% and 98.48%, respectively, with leave-1-out and bootstrapping validations.

CONJUGATE GRADIENT AND APPROXIMATE NEWTON METHODS FOR AN OPTIMAL PROBABILISTIC NEURAL NETWORK FOR FOOD COLOR CLASSIFICATION

The probabilistic neural network (PNN) is based on the esti- mation of the probability density functions. The estimation of these den- sity functions uses smoothing parameters that represent the width of the activation functions. A two-step numerical procedure is developed for the optimization of the smoothing parameters of the PNN: a rough optimiza- tion by the conjugate gradient method and a fine optimization by the approximate Newton method. The thrust is to compare the classification performances of the improved PNN and the standard back-propagation neural network (BPNN). Comparisons are performed on a food quality problem: french fry classification into three different color classes (light, normal, and dark). The optimized PNN correctly classifies 96.19% of the test data, whereas the BPNN classifies only 93.27% of the same data. Moreover, the PNN is more stable than the BPNN with regard to the random initialization. The optimized PNN requires 1464 s for training compared to only 71 s required by the BPNN.

Olfactory receptor-based polypeptide sensor for acetic acid VOC detection

Rapid detection of food-borne pathogens in packaged food products can prevent the spread of infectious diseases. This study investigates the application of novel sensing material that is sensitive to specific indicator volatile organic compound (VOC) related to Salmonella contamination in packaged meat. Specifically, the objective was to develop an olfactory receptor-based synthetic polypeptide sensor for the detecting acetic acid in low concentrations and at room temperature. Synthetic polypeptide was deposited on a quartz crystal microbalance (QCM) electrode and was evaluated for detecting acetic acid at 10-100 ppm. Developed sensor exhibited repeatable response to a particular concentration of acetic acid and displayed reproducibility among multiple sensors during acetic acid detection. Mean estimated lower detection limits of these sensors were about 1-3 ppm and linear calibration models showed linear relationships. Thus, the QCM sensors exhibited a great potential for detecting low concentrations of acetic acid at room temperature and can be used in the sensor array for packaged meat spoilage and contamination detection.

Background segmentation and dimensional measurement of corn germplasm

An automatic thresholding technique was developed to segment the background from the images of corn germplasm (ears of corn). The technique was a modification of Otsu’s algorithm using probability theory. Three different measures were used to evaluate the performance of the modified Otsu’s algorithm for background segmentation and subsequent dimensional measurement of corn germplasm. Modified Otsu’s algorithm was found to perform better than Otsu’s algorithm and was successful in automatic background segmentation of all 80 images of corn germplasm included in the study. This modified algorithm also eliminated the misclassification of exposed cob in the image as background which occurred with Otsu’s algorithm. Subsequent dimensional measurements based on the segmentation by the modified algorithm were also highly accurate.

Evaluations of fractal geometry and invariant moments for shape classification of corn germplasm

Abstract Computer vision-based techniques were developed and evaluated for classifying different shapes of germplasms (ear of corn). An algorithm was developed to discriminate round-shaped germplasms based on two features, i.e. circularity and dimensional ratio. Two different approaches based on fractal geometry and higher order invariant moments were used for classification of non-round shaped germplasms. In the fractal-based approach, two additional fractal geometry-based features (i.e. fractal-shape factor and fractal perimeter) were developed and used with fractal dimension and aspect ratio to represent the shape features of the germplasms. Classifications rules based on modified Euclidean measures and distance weighted K -nearest neighborhood were used for classifying the germplasms into one of three non-round-shape classes (cylindrical, cylindrical-conical and conical). Though the overall correspondence for classifying non-round germplasms was 60% (based on 80 samples), a maximum correspondence of 80% could be obtained for classifying cylindrical germplasms (based on 18 samples). Neither method could provide similar classification correspondence for cylindrical-conical germplasms. On the other hand, these methods, however, showed a correspondence of 82.5% for classifying non-round corn germplasms into cylindrical and non-cylindrical (conical and cylindrical-conical) shapes.

STUDY OF HEADSPACE GASES ASSOCIATED WITH SALMONELLA CONTAMINATION OF STERILE BEEF IN VIALS USING HS-SPME/GC-MS

Sterile beef (fresh strip loins) samples were inoculated with Salmonella typhimurium, and both control and inoculated samples were stored at 20°C in 20 mL headspace vials covered with food-grade cling film. An array of volatile compounds was detected in the headspace of the control and inoculated samples. The study was conducted for four days, and the volatiles in the headspace were analyzed each day using manual headspace solid-phase microextraction (HS-SPME) in combination with gas chromatography-mass spectrometry (GC-MS). Acetic acid, ethanol, carbon dioxide, and 3-hydroxy-2-butanone were the most prominent compounds detected in the study. The F-tests (Fishers variance ratio) for the main effect of the sample source established acetic acid and ethanol as compounds of interest for monitoring the status of Salmonella in raw fresh beef. Good linear correlations were found between the logarithms of the peak area responses of these compounds with Salmonella count.

Computer-based neuro-vision system for color classification of french fries

French fries are one of the frozen foods with rising demands in domestic and international markets. Color is one of the critical attributes for quality evaluation of french fries. This study discusses the development of a color computer vision system and the integration of neural network technology for objective color evaluation and classification of french fries. The classification accuracy of a prototype back-propagation network developed for this purpose was found to be 96%.

Neural-Network-Based Classification of Meat: Evaluation of Techniques to Overcome Small Dataset Problems

One of the objectives of our multidisciplinary research group is to develop sensors for the detection of Salmonella contamination in beef. Similar to most biological studies, beef contamination classification studies using artificial neural networks (ANNs) are restricted to small datasets. This study evaluates selected techniques of data domain expansion and synthetic sample generation on small datasets associated with meat contamination. Mega-trend-diffusion (MTD) and functional virtual population (FVP) techniques for data domain expansion and synthetic sample generation were assessed on the small datasets. The datasets used were obtained from a thin-film (TF) module electronic nose system in response to the headspace of control and Salmonella-inoculated packaged meat samples. Back-propagation neural networks (BPNNs) were used to determine classification accuracies of the synthetically expanded datasets. For aged beef datasets, the maximum mean of average overall classification accuracies provided by FVP technique was 90%. The maximum mean of average overall classification accuracies obtained by FVP technique was about 81% for fresh beef datasets. MTD technique also provided similar accuracies (in the lower 80s). Both techniques were found useful for expanding the domain range of the small dataset in order to test and evaluate BPNN-based classification models.