A feature dimensionality reduction strategy coupled with an electronic nose to identify the quality of egg

Abstract

In this study, a feature dimensionality reduction strategy is proposed to reduce the feature dimensionality of the electronic nose (e‐nose) sensor, combined with support vector machine (SVM) to distinguish the gas information of eggs produced by chickens with different breeding methods. First, to characterize the overall properties of the original detection signal, five different time domain features are extracted from each sensor. Second, max‐relevance and min‐redundancy (MRMR) is introduced to obtain a preliminary optimal feature set. Finally, kernel principal component analysis (KPCA) is introduced to further eliminate the correlation between features and obtain the optimal feature set. The result shows that the optimal feature set is obtained by MRMR–KPCA, and good classification accuracy is obtained based on SVM. In conclusion, the feature dimensionality reduction strategy effectively reduces the feature dimensionality of the e‐nose sensor, eliminates the correlation between features, realizes the nondestructive detection for the quality of egg, and provides an effective technical method for the market quality supervision of egg. Practical applications In different breeding conditions, the nutritional value of eggs produced by chickens is different. To get more benefit, some inferior eggs are brought into the market instead of those with a higher nutritional value. Therefore, it is very important to use the nondestructive detection technology to quickly identify the quality of egg. In this work, e‐nose is used to obtain the gas information of eggs produced by chickens with different breeding methods. A feature dimensionality reduction strategy is proposed to process the e‐nose data, which realizes the effective identification of gas information of eggs. Moreover, it provides an effective detection method for the quality monitoring of the egg market.

Identification of Panax Notoginseng Powder in Different Parts Based on the Electronic Nose and Time-Domain Feature Extraction

“…The device consists of a metal oxide gas sensor array consisting of 10 different MOS sensors, a gas sampling device, and a signal processing unit. 30 The structure of the device is shown in Fig. 1, and the performance comparison of the gas sensor array is listed in Table I.…”

A machine-learning method isolating changes in wrist kinematics that identify age-related changes in arm movement

“…The Minimal Redundancy Maximal Relevance (MRMR) algorithm first introduced by Peng et al 31 is a powerful algorithm that identifies a subset of input features (independent variables) by simultaneously maximizing relevance to the target variable and minimizing redundancy among selected features. Other feature selection methods in ML studies have used standalone methods such as analysis of variance 32 – 37 or a two-step process combining the MRMR algorithm with secondary techniques such as with Fischer scores 38 , kernel canonical correlation analysis 39 and kernel principal component analysis 40 . However, we are not aware of any studies that have used the MRMR algorithm followed by a variance threshold as a feature selection technique for an SVM to distinguish PT in young and older adults.…”

Beef Quality Classification with Reduced E-Nose Data Features According to Beef Cut Types

“…As food spoils, odor and gas are emitted due to microbial activity [ 5 ]. Here, the odor is caused by the different gases that spread, and it is possible to obtain information about the quality of food by detecting these emitted gases through sensors [ 6 , 7 ]. These systems, which contain a sensor array and imitate the human nose, are called electronic noses (e-noses).…”

“…There are many studies in the literature that use data from e-noses to automatically determine the quality of different foods [ 7 , 12 , 13 , 14 ]. E-nose studies tracing perishable foods are more common [ 15 , 16 ].…”