Application of Electronic Nose, Spectroscopy Sensor, and Machine Learning for the Freshness Evaluation of Fish Fillet
學生姓名:
廖玉芊
指導教授:
方銘志
學期:
110下
摘 要:
Biochemical, microbiological, sensory and oxidation-reduction reactions are associated with the deterioration of fish quality during trading, transiting and storage. The estimation of freshness is an important issue in fish which is related to its overall quality. In the past, a large number of different measurement methods were applied to measure the freshness of fish. However, many of them required complex sample preparation, experienced operators, or high cost of instruments. In recent years, metal oxide semiconductor (MOS) and machining learning have been widespread used in food quality evaluation. In our research, ten gas sensors and a multi-wavelength spectroscopy sensor, AS7265x, were designed for the freshness detection of fish during storage period. The performance of designed gas sensors system was evaluated. The results showed the response and recovery time was fast, and all in 5 minutes. Among all sensors, MQ-138 sensor observed an even shorter response in 1 minute. The linearities of MQ sensors were tested by simulating a concentration-dependent gas phase environment through different aqueous
alcohol solutions. The coefficient of determination (R2 ) of MQ-3 and MQ-138 sensors were over 0.99 which offered great response linearity. The other sensors also showed good linearities with R2 at least over 0.95. The sensor-to-sensor variability was also tested by comparing four of the identical MQ sensors in different concentrations of alcohol aqueous solution. However, significant varieties were obtained between identical MQ sensors. Therefore, two of each identical sensors with closed response were selected and assembled into two sensor boxes for further application of fish freshness fast detection. Currently, the sensor abilities for using as volatiles detection units in the designed sensor boxes was confirmed. The future studies will focus on the detection of 50 fillets by the designed sensor box, and comparison of sensor results with the measured TVB-N and K value during storage. The obtained date will be processed using artificial neural network (ANN) model to build a fish freshness prediction system.
alcohol solutions. The coefficient of determination (R2 ) of MQ-3 and MQ-138 sensors were over 0.99 which offered great response linearity. The other sensors also showed good linearities with R2 at least over 0.95. The sensor-to-sensor variability was also tested by comparing four of the identical MQ sensors in different concentrations of alcohol aqueous solution. However, significant varieties were obtained between identical MQ sensors. Therefore, two of each identical sensors with closed response were selected and assembled into two sensor boxes for further application of fish freshness fast detection. Currently, the sensor abilities for using as volatiles detection units in the designed sensor boxes was confirmed. The future studies will focus on the detection of 50 fillets by the designed sensor box, and comparison of sensor results with the measured TVB-N and K value during storage. The obtained date will be processed using artificial neural network (ANN) model to build a fish freshness prediction system.
專題討論20220504廖玉芊摘要.pdf