Research On Novel Detection Techniques For Fish Edible Safety Based On Taste Sensors

Fish is an important part of human diet. In recent years, fish production has been increasing year after year in China. Meanwhile edible safety of fishes is also drawing extensive attentions from the public gradually. The microbial contamination, heavy metals residues and fish drugs residues are the main hazard factors influencing fish edible safety. Conventionally, fish microbial contamination is detected using plate count method, and heavy metals residues and fish drugs residues are determined based on physical and chemical analysis. However, the plate count method is time-consuming and requires highly complex operation steps; while, the physical and chemical analysis methods are generally high cost, time-consuming and need inconvenient operations. Hence, it is very meaningful to develop rapid and convenient methods for fish microbial contamination detection and low-cost and convenient methods for heavy metals residues and fish drugs residues determination. In this study, novel taste sensing technologies-electronic tongue technique and taste visualization technique-combined with pattern recognition methods were proposed to detect the fish microbial contamination, heavy metals residues and fish drugs residues for quality guaranteed. The main contents are as follows: 1. Study on novel detection methods for fish microbial contamination and fish drugs residues based on electronic tongue techniqueEstablishment of the novel detection method for fish microbial contamination based on electronic tongue. Fish species of Carassius auratus was used as the samples. The standard plate count method was used to detect the total viable counts (TVC) in fish samples stored at 4℃ in different days as the fish microbial contamination index. The electronic tongue was employed to detect the homogenates of fish and deionized water. Fisher discriminant analysis, Bayes discriminant analysis and distance discriminant analysis were applied comparatively to build recognition models for qualitatively analysis of fish microbial contamination. Results show that, the identification rates in training set and prediction set of the Fisher discriminant analysis model were 92.71% and 85.44% respectively; the identification rates in training set and prediction set of the Bayes discriminant analysis model were 100% and 93.75% respectively; the identification rates in training set and prediction set of the distance discriminant analysis model were all 100%. It is can be observed from the results that the distance discriminant analysis model was the best model for fish microbial contamination discriminant. The partial least squares (PLS), support vector machine (SVM) and BP-artificial neural network (BP-ANN) were applied comparatively to build models for fish TVCs prediction with electronic tongue sensors as the input variables and the TVC as the output variable. Results show that, the correlation coefficient in the prediction set (Rp) of the PLS, SVM and BP-ANN models were 0.859,0.9683 and 0.993 respectively. The root mean square errors in the prediction set (RMSEP) of the PLS, SVM and BP-ANN models were 0.927 In CFU/g,0.516 In CFU/g and 0.217 1n CFU/g respectively. The SVM and BP-ANN models were better than the linear PLS model.Establishment of the novel detection method for fish drugs residues based on electronic tongue. The high performance liquid chromatography (HPLC) method was employed to detect fish drugs residues of fluoroquinolones (enrofloxacin and sarafloxacin) in fish samples and the outcomes were used as the reference values for electronic tongue measurements. The phosphate buffer solution was selected to soak the drugs in fish as the liquid samples detected by electronic tongue. The SVM and BP-ANN were used to build models for fish drugs residues prediction. Results show that, Rps of the SVM models for prediction of the enrofloxacin and sarafloxacin residues in fish samples were 0.732 and 0.843 respectively, and the RMSEPs were 23.77 μg/kg and 3.478 μg/kg respectively; Rps of the BP-ANN models were 0.816 and 0.844 respectively, and the RMSEPs were 16.64 μg/kg and 4.118 μg/kg respectively. The performances of the BP-ANN models for fish drugs residues prediction were better than the SVM models.2. Creation of the novel detection method for fish heavy metals residues and fish drugs residues based on taste sensorsThe mechanism of color changes of pyridylazo compounds and porphyrin compounds reacted with metal ions by coordination interaction was studied. Afterwards, detection method based on the novel taste sensing mechanism was developed for fish heavy metals residues. Five pyridylazo compounds and four porphyrin compounds were selected as the sensitive materials for constructing the taste visualization sensors used to detect heavy metals residues in fish. The (Inductively coupled plasma mass spectrometry) ICP-MS method was employed to detect the heavy metals (Pb, Cd, Hg) in fish samples and the measurement results were used as the reference for taste visualization measurements. Meanwhile, the weak acid liquid solutions obtained from the microwave digested samples adjusted by sodium hydroxide were detected using the novel taste visualization sensors. The SVM and BP-ANN were applied respectively to build models for fish heavy metals residues prediction. Results show that, the Rps of the BP-ANN models for Pb, Cd, Hg prediction were 0.807,0.811and 0.865 respectively, and the RMSEPs were 0.133 mg/kg,0.023 mg/kg and 0.017 mg/kg respectively; the Rps of the SVM models for Pb, Cd, Hg prediction were 0.745,0.74 and 0.718 respectively, and the RMSEPs were 0.107 mg/kg,0.013 mg/kg and 0.023 mg/kg respectively. The performances of the BP-ANN models were better than the SVM models.The novel taste visualization sensors array was constructed for fish drugs residues detection. Four porphyrin compounds, seven metal porphyrin compounds and one metal phthalocyanine compound were selected as the sensitive materials to construct the taste visualization sensor arrays for fish drugs residues detection. The sensitive materials selected was according to the color changes of the metalloporphyrin compounds reacted with organic micro-molecular compounds via the intermolecular forces of coordination, hydrogen bond, π-π stacking, etcetera. The results of HPLC measurements for enrofloxacin and sarafloxacin residues in fish samples were used as the reference data for taste visualization measurements. The leach liquors samples of the fish drugs extracted by phosphate buffer solution were detected using the novel taste visualization sensors. The SVM and BP-ANN were applied respectively to build models for fish drugs prediction. Results show that, the Rps of the SVM models for enrofloxacin and sarafloxacin prediction were 0.778 and 0.746 respectively, and the RMSEPs were 18.61 μg/kg and 6.43 μg/kg respectively; Compare with the SVM models, the Rps of the BP-ANN models for enrofloxacin and sarafloxacin prediction were improved to 0.83 and 0.779 respectively, and the RMSEPs were 12.98 μg/kg and 4.913 μg/kg respectively.3. Development of the intelligent prediction method for fish drugs residues based on multi-sensor information fusionThe intelligent prediction method for fish drugs residues was developed based on multi-sensor information fusion technique by fusing the electronic tongue data and taste visualization sensors information. The phosphate buffer solution was used to soak the fluoroquinolones drugs in fish as the liquid samples detected by electronic tongue and taste visualization sensors respectively. The linear function transformation method was used to normalize combined data of the characteristic variable information of the electronic tongue sensors and taste visualization sensors. The SVM and BP-ANN models were built for fish drugs residues prediction based on feature level fusion technique with the HPLC measurements results as the reference. Results show that, the Rps of the SVM models for enrofloxacin and sarafloxacin prediction were 0.867 and 0.87 respectively, and the RMSEPs were 17.09 μg/kg and 3.33 μg/kg respectively; The Rps of the BP-ANN models for enrofloxacin and sarafloxacin prediction were 0.852 and 0.864 respectively, and the RMSEPs were 13.28 μg/kg and 4.83 μg/kg respectively. It is can be observed from the results that the multi-sensor fusion of the electronic tongue and taste visualization technique was better than each single one of them for prediction of the fish drugs residues in fish samples.All the results demonstrated that:the novel detection method established based on electronic tongue could be used for fish microbial contamination discriminant and quantitative prediction of the TVC and drugs residues in fish; the novel detection method created based on taste sensors could be used for quantitative prediction of the fish heavy metals residues and fish drugs residues; the performance of the intelligent prediction method developed based on multi-sensor information fusion for fish drugs residues was better than each one of the electronic tongue and taste visualization technique. All the results of the study show that the taste sensing technology has a huge potential in detection of the fish edible safety.