| Salmon Salmon,as a meat food with delicious meat and rich nutritional value,is popular worldwide.In recent years,the steady development of the global economy has brought people a better living environment.At the same time,the consumption capacity has also risen sharply,resulting in a sharp increase in the demand for salmon.China’s salmon mainly depends on imports,and the transportation process is mainly based on freezing and cold storage.However,compared with other meat,fish meat has relatively more water content,which provides a suitable parasitic environment for microorganisms.In addition,with long time of transportation,meat quality is prone to decay and deterioration in the environment.Compared with frozen salmon,frozen salmon can better retain its freshness,making it more similar to the taste and meat quality of fresh salmon.In the process of selling,it is difficult for consumers to distinguish the quality of frozen and frozen salmon with the naked eye.Many illegal traders freeze and sell the poor quality of frozen salmon to make higher profits.Therefore,the quality and safety testing of salmon is becoming increasingly important.In response to the above issues,this article adopts spectral and hyperspectral imaging techniques,combined with various stoichiometric methods and image processing techniques,to conduct stoichiometric analysis on frozen and refrigerated salmon meat,and establishes predictive models for qualitative analysis.At the same time,visualization research on salmon fat content is conducted based on MCR-ALS hyperspectral reconstruction technology.The above experimental results indicate that VIS-NIR spectroscopy and hyperspectral imaging technology have good feasibility in detecting the quality of salmon.At the same time,this article also conducted research on the design of portable instruments for rapid measurement of salmon quality based on spectral technology,providing a certain technical solution and theoretical basis for the development of digital fisheries.The main content and conclusions of the study include:(1)Based on near-infrared spectroscopy and hyperspectral imaging technology,the quality identification of fresh and different frozen salmon samples was studied.Taking fresh salmon and salmon under different freezing and thawing days as experimental samples,select the region of interest in ENVI 4.6 to extract the average spectral data from the collected hyperspectral images,and repeat the operation on all samples to obtain the sample spectral data matrix.Then,the convolution neural network(CNN)model is established by comparing the extracted spectral data with the original spectral data,and finally it is concluded that the1-Der-CNN model established by the data pretreated by the first derivative has the best prediction accuracy,and the correct recognition rate of the five prediction samples can reach97.92%.At the same time,in order to reduce the data dimension,the characteristic wavelengths of CARS and SPA are extracted on the basis of 1-Der data.In order to verify the reliability of the model,three prediction models based on characteristic wavelengths,CNN,BP-ANN and LS-SVM,are established for comparative experiments,and the optimal prediction model is selected.From the results of the model,we can see that the CNN model based on the two feature wavelength extraction has achieved better correct recognition rate,and the correct recognition rate of the best model CARS-CNN is 93.75%.(2)Based on near-infrared hyperspectral imaging technology,the rapid identification and visualization of salmon in different cold storage days were studied.First,the purchased salmon samples were subjected to cold storage experiments under different days,and then the corresponding hyperspectral images were extracted from the corresponding cold storage time samples,and the average spectral data were extracted from the regions of interest of the images,and the PLSR,BPNN and LS-SVM prediction models of the full spectral data were established.At the same time,in order to test whether the relatively small number of characteristic wavelengths with high representativeness also have high prediction accuracy,the 24 characteristic wavelengths obtained after the feature screening of RF data bands are used for comparative experiments,and three prediction models with full spectrum data are established for comparative analysis.The RF-LS-SVM model has the highest prediction accuracy,with the modeling set determination coefficient(R2cal)and root mean square error(RMSEC)of 0.9343 and 1.0130,and the prediction set determination coefficient(R2pre)and root mean square error(RMSEP)of 0.9237 and 1.0217,respectively.Finally,the optimal model is used to predict that each pixel point in the hyperspectral image of the sample corresponds to different cold storage time,and the image is programmed and processed in the MATLAB software.The pixel points of the image with different cold storage time are expressed in different colors,and the pseudo-color image of the image coordinate information is constructed to realize the visual identification of salmon under different cold storage days.(3)Use MCR-ALS hyperspectral reconstruction to realize salmon fat visualization.First,the original spectral data is reconstructed by MCR-ALS to obtain the reconstructed spectral data,and the SPA algorithm is used to extract the characteristic wavelength of the original data and the reconstructed data to reduce the complexity of subsequent modeling and reduce the time of model calculation.Then,through the establishment of LS-SVM model,from the comparison of models,the correlation coefficient of the full spectrum(R=0.9657)and the correlation coefficient of the characteristic band(R=0.9555)are better than the original spectral data.The hyperspectral image can be well visualized by bringing it into the optimal model of MCR-ALS-SPA-LS-SVM.In the fat visualization image,the fat strip distribution is more clear and accurate,and the edge contour information is corrected,showing excellent visualization potential.It also shows that MCR-ALS plays an important role in improving image spatial model and reducing background constraints.(4)A development plan and implementation process for a portable instrument for rapid measurement of salmon quality using near-infrared spectroscopy were proposed.At the same time,the performance parameters of the instrument’s core components and key points for attention during the experimental process were compared and analyzed.A simple basic design of hardware communication and software interface was completed.Using salmon samples to measure fat content,the instrument was validated and analyzed.Finally,through data comparison,it was found that the deviation curve of the portable instrument for rapid measurement of salmon fat content fluctuated greatly,and the measurement accuracy was not high enough.However,it also had a certain detection effect.Further improvements are needed to the portable instrument in the future to improve measurement accuracy and applicability,and achieve the goal of rapid detection. |
PDF Full Text Request