Research And Application Of Freshness Recognition Of Cold And Fresh Yellow Croaker Based On The Residual Neural Network Model

Large yellow croaker（Pseudosciaena crocea）is an important economic fish in China,due to its delicious taste and abundant nutrition.However,the freshness of large yellow croakers is easy to change.Therefore,the detection of the freshness of large yellow croakers is important.Traditional detection methods of freshness are professional,time-consuming,and difficult.To ensure the integrity of samples,it is important to develop a fast,accurate,and non-destructive detection method to identify the freshness of large yellow croakers.The machine vision system can simulate the human eye to collect images.This method is not affected by subjective factors of human eye evaluation and does not touch samples.The deep learning model,represented by the residual neural network,can quickly obtain classification results with high accuracy.In this study,the correlation between freshness the change rule and sensory evaluation was studied.The Spearman correlation analysis method was used to screen out the freshness indexes that easily affect sensory evaluation.Furthermore,the images of large yellow croakers at different storage temperatures were collected by a machine vision system,and the freshness levels and classified as Class I and Class II（the Chinese National Standards）by a 34-layer residual neural network（ResNet-34）model.In addition,the application program for chilled large yellow croakers’image identification was developed to obtain the freshness classification online through a simple operation.The main research results are as follows:Firstly,the relationship between the freshness indices（p H,TVC,TBA,TMA,TVB-N）and a sensory score of large yellow croakers at different storage temperatures（0℃,4℃,10℃,15℃,25℃）was studied.Spearman correlation analysis showed that the correlation coefficient between sensory evaluation and TVB-N was the highest,as the correlation coefficients were0.987,0.978,and 0.984 at 4℃,10℃,and 25℃,indicating that TVB-N was the main factor affecting sensory evaluation.In the following experiment,we used TVB-N to classify the freshness of large yellow croakers.Secondly,a freshness identification model of large yellow croaker based on ResNet-34was constructed.TVB-N classification-based images at different storage temperatures were divided into Class I and Class II（the Chinese National Standards）by using the model analysis.The results showed that the recognition accuracy of the model at different storage temperatures（0℃,4℃,10℃,15℃,25℃）was above 0.85,suggesting that the model can accurately identify the freshness of large yellow croaker.To further evaluate the effectiveness of the model accuracy,the parameter F₁-score was evaluated.F₁-score of large yellow croakers at different storage temperatures were all greater than 0.80（except for the identification of the Class I freshness of mixed whole,F₁-score 0.76）.These results proved that ResNet-3 has a good performance in identifying different freshness images of large yellow croakers.In addition,the visual analysis of the model showed that the model extracts the key information in the image and ignores the background information.Finally,an application program for freshness image recognition of chilled large yellow croakers based on ResNet-34 was developed.The modules of the program（access module,image capture module,a freshness identification module,and result display module）were designed in Java language.The test results of the program showed that the recognition accuracy of the whole,gill,and body can reach 90%,92%,and 86%,respectively.The results showed that the program can effectively identify different freshness images of large yellow croakers.Users can obtain the freshness grade of large yellow croakers by a simple operation.