1. Introduction
Growing demand for nutritious, high-quality, and safe food has made the food industry and consumers pay more attention. Food adulteration is the biggest problem in the food supply chain and is prevalent at various levels. Food adulteration, committed wittingly or unwittingly, is a major problem and can have severe consequences for people with food intolerances, lifestyles, and religious practices. The most common reasons for food adulteration entail substituting high-value food products with low-value food products for financial profits [
1,
2]. Beef is one of the most widely consumed of all meats because it is low in fat, high in protein, vitamins, amino acids, and other essential nutrients required for human beings [
3]. According to the United States Department of Agriculture’s (USDA) world market and trade reports, the United States is anticipated to be the world’s top producer and exporter of beef in 2022 [
4]. Because of its high consumption and economic benefits, beef is the prime target for adulteration and is most frequently adulterated with pork due to its lower economic value. Consumers object to such adulterations for allergic, cultural, monetary, religious, and consumer rights reasons [
5,
6]. In an investigation conducted in China on 1553 media reports, of the total instances of adulteration, animal-based food makes up 37.78% [
7]. Halal-certified meat pies served to prisoners in UK jail in 2013 were found to contain pork DNA in them [
8]. With the horse meat scandal incident in 2013, the demand for beef authentication has increased significantly [
9]. With these types of incidents, there is a rise in the need for an accurate and reliable method of detecting beef adulteration.
To detect beef adulteration, various techniques are employed such as molecular biology-based methods [
10,
11,
12,
13,
14,
15,
16], spectroscopy methods [
3,
6,
17,
18,
19,
20,
21], chromatographic methods [
22,
23,
24], and enzyme-linked immunological methods [
25,
26,
27]. However, the equipment to implement such techniques is expensive, poor in real time, time consuming, destructive, laborious, and requires technical supervision. Additionally, the equipment requires highly skilled individuals to conduct the experiment and analyze the results. This demands the development of an instrument that is simple, accurate, economical, portable, quick, highly sensitive, and requires fewer samples for measurement.
An electronic nose (E-nose) is a device that mimics a human’s sense of smell. It is simple, sensitive, reliable, low-priced, non-destructive, and highly correlated with the detection of various odor signatures. An E-nose is built up of an array of sensors, signal processing units, and a pattern recognition system to measure the testing sample. The sensors that are used for the design of a sensor array are metal-oxide semiconductor (MOS) gas sensors [
28], metal-oxide semiconductor field effect transistors (MOSFET) gas sensors [
29], acoustic wave gas sensors [
30], quartz crystal micro balance sensors [
31], infrared sensors [
32], colorimetric sensors [
33], fluorescence sensors [
34], conducting polymer gas sensors [
35], and fiber optic gas sensors [
36], among others. Data from the sensor array are received by the signal processing unit and pre-processes the data as per the requirement. A pattern recognition system performs the classification or regression of the processed data to measure the sample. The development at all stages of E-nose design made it a good alternative to traditional methods. E-nose is employed in different food industries such as the meat industry [
37,
38], oil industry [
39], alcohol industry [
40], tea industry, [
41] etc. Furthermore, the E-nose is used in various food industry applications such as adulteration detection [
42], origin identification [
43], authentication [
44,
45], grading [
46,
47], freshness assessment [
40], and shelf-life determination [
48]. In this study, we develop a smart E-Nose (SE-Nose) for the qualitative and quantitative analysis of pork adulteration in beef. The smartness of the E-Nose is achieved by introducing a sample slicing window protocol (SSWP) for fast-track detection, designing a less complex system, qualitative analysis of adulteration using artificial intelligence-based classification models, and the quantitative analysis of adulteration using regression models, and achieving higher accuracies. In other parts of this section, an extensive literature analysis of evaluating the E-nose to identify beef adulterated with pork is described.
E-nose, composed of ten semi-conductive polymer sensors with multi-discriminant analysis, is used to identify beef–pork mixtures in four compositions [
49]. The cooked and uncooked samples are stored at 4 °C for 0–6 days. The sensing time for measuring each sample is of 4 min. The presented results clearly show that the designed E-nose identifies samples according to mixture composition and storage days. FOX4000 E-nose with discriminant factor analysis (DFA), principal component analysis (PCA), and partial least square analysis (PLS) techniques are used to detect pork adulteration in beef [
50]. The results demonstrate that the system is capable of detecting pork adulteration in beef in various ratios. The coefficient of determination (R2) between sensor data and minced pork ratios is 0.9762. To study and find the optimal temperature required for the measurement sample to be tested using E-nose for discriminating beef–pork mixtures, a sensory array made of eight MOS sensors with temperature and humidity sensors is designed [
45,
51]. Five beef–pork mixture classes are discriminated in this study using a support vector machine (SVM), k-nearest neighbors (KNN), random forest, and naive Bayes techniques at −22 °C, room temperature, and 55 °C, respectively. According to this study, continuous gases from beef and pork can be detected at −22 °C with maximum accuracy to detect beef adulterated with pork. Using the same sensor array, another author identified seven beef–pork mixtures with an optimized E-nose system designed using a sensor array followed by noise filtering and optimized SVM blocks [
42]. Each mixture sample is measured for 15 min, and 60 data points are recorded for each measurement. An accuracy of 98.10% is obtained with the designed system. To analyze the effect of gas concentration in the detection and classification of beef–pork mixture, an E-nose consisting of nine MOS sensors is used [
52]. The proportion of beef–pork mixture used is the same as that used by [
42]. Sample sizes of 50 mL, 150 mL, and 250 mL is used for measurement, and each sample is measured for 2 min. Statistical feature extraction, classification, and ensemble learning are implemented on the measured data to determine the system’s performance. Good classification values are obtained using the ensemble learning method with 50 mL. An E-nose designed with an array of colorimetric sensors with twelve chemical dyes was developed to detect beef adulterated with pork [
33]. The twelve chemical dyes of the colorimetric sensor array are selected with chemically receptive dyes that can detect complex volatile organic compounds (VOCs) generated from meat samples. The colorimetric sensor array is placed in the sample chamber for five minutes so that the dyes on the sensor array react with the VOC from the meat sample. The classification methods Fisher linear discriminant analysis (Fisher LDA) and extreme learning machine (ELM) are used to classify the three mixture combinations of pork adulteration in beef. The ELM method yields an accuracy of 87.5% with the prediction set. Additionally, the adulteration level is calculated using a backpropagation artificial neural network (BP-ANN), and a correlation coefficient (R) and root mean square error (RMSE) of 0.85 and 0.147 are obtained, respectively.
It is clear from the literature discussed above that the E-nose can predict beef adulterated with pork. The problems identified in the literature are the longer measurement time, low classification accuracy, lesser beef–pork mixture ratios, the complex design of architecture, and the missing quantitative analysis. This study developed the SE-Nose methodology for the qualitative and quantitative fast-track detection of pork adulteration in beef. The implementation of SE-Nose is performed in four stages: data collection, pre-processing, pattern recognition, and output. The data collection stage deals with the dataset used to validate the methodology, and the pork adulterated with beef dataset available in the literature [
53,
54] was used in our study. This dataset helps to overcome the problem of a less beef–pork mixture ratio identified in the literature, as this dataset contains seven combinations of beef–pork mixture ratios. The pre-processing stage contains SSWP and normalization blocks. The SSWP extracts the early part of the signal to achieve fast-track detection, thereby overcoming the problem of a longer measurement time and complex architecture problems identified in the literature. In the pattern recognition stage, multiple classifications and regression models are implemented for the classification and identification of adulteration levels present in the sample. With the design parameters of pattern recognition models (SVM, ANN), the problems of low-classification accuracy and complex architectures are solved. The output stage publishes the qualitative and quantitative analysis of sample adulteration using the results obtained from the pattern recognition model. The performance of the SE-Nose system is evaluated using the size of input data, the count of windows, recognition time, training time, validation time, accuracy, mean square error (MSE), RMSE, R, and R2. Furthermore, the results obtained with the SE-Nose system are compared with the results present in the available literature.
The organization of the remaining sections of the paper is as follows:
Section 2 discusses the SE-Nose methodology consisting of pork adulteration in the beef dataset, SSWP, normalization, classification models, regression models, and outputs.
Section 3 discusses the results obtained with SSWP, classification models, and regression models for the qualitative and quantitative analysis of pork adulteration in beef. Furthermore,
Section 3 compares the results obtained with the SE-Nose methodology with comparable results available in the literature.
Section 4 presents the conclusion of our study.