Premium
Nondestructive prediction modeling of S‐ovalbumin content in stored eggs based on hyperspectral fusion information
Author(s) -
Fu DanDan,
Wang QiaoHua,
Ma MeiHu,
Ma YiXiao,
Wang Bin
Publication year - 2019
Publication title -
journal of food process engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.507
H-Index - 45
eISSN - 1745-4530
pISSN - 0145-8876
DOI - 10.1111/jfpe.13015
Subject(s) - hyperspectral imaging , partial least squares regression , ovalbumin , biological system , computer science , feature (linguistics) , content (measure theory) , pattern recognition (psychology) , artificial intelligence , fusion , mathematics , machine learning , biology , mathematical analysis , linguistics , philosophy , immunology , immune system
Hyperspectral imaging technology was used to nondestructively predict S‐ovalbumin content by obtaining transmission spectral and fused image data. Hyperspectral images of eggs with spectral wavelengths between 300 and 1,100 nm were collected. Three image features were selected to use as variables to model. After spectral pretreatment, 14 characteristic wavelengths were selected using competitive adaptive reweighted sampling algorithm (CARS). Nondestructive prediction models of S‐ovalbumin content were developed based on either image feature parameters, a combination of spectral characteristic wavelengths, or image‐spectral fusion data after dimensionality reduction. A partial least squares (PLS) model and genetic algorithm to optimize a back propagation neural network (GA‐BP) model was established based on these three kinds of data. From comparisons between PLS and GA‐BP models based on the same type of each of the three data types, we determined that hyperspectral fusion data and a GA‐BP model can improve the performance of a S‐ovalbumin nondestructive prediction model. It had a simpler structure with fewer variables. So, it can be used to nondestructively and rapidly predict protein content in egg white. Practical applications Current nondestructive methods to evaluate egg quality primarily focus on the changes of simple external and internal quality, such as egg weight and simple chemical indicators. Due to the individual differences between eggs, the above methods are under a limited application scope. Changes in S‐ovalbumin content do not depend on egg weight, and so forth, individual differences have little effect on it. Changes in S‐ovalbumin content of various varieties of eggs are alike because the biochemical processes that affect egg white protein content is fundamentally the same across egg varieties, and its traditional detection method is to perform complex biochemical tests after breaking. Therefore, if the egg's internal protein content can detected quickly without breaking it, it is more accurate and more profound from the biochemical essence and it can not only solve the shortcomings of biochemical detection, but also provide important theory and application technology for the detection and classification of egg quality.