Skip to main content
NCBI home page
Poultry Science logoLink to Poultry Science
. 2025 Mar 8;104(4):104999. doi: 10.1016/j.psj.2025.104999

Prediction of chicken breast meat freshness based on hyperspectral imaging technique and high-throughput sequencing

Xingyu Shen a, Lujuan Xing a,1, Leiqing Pan b, Yujia Miao a, Wangang Zhang a,⁎,1
PMCID: PMC11951181  PMID: 40081174

Highlights

  • •

    The quality indicators (color, pH, TVC, TVB-N and TBARS) were detected to reflect trends in freshness of chicken breast meat during storage at 4°C.

  • •

    Three dominant spoilage microorganisms (Pseudomonas, Brochothrix and Escherichia) were screened by high-throughput sequencing in chicken breast meat.

  • •

    Predictive modeling of TVC, Pseudomonas, Brochothrix and Escherichia content were compared based on visible near-infrared (400-1000 nm) hyperspectral imaging technique.

  • •

    Finally, the Pseudomonas count was chosen as the characteristic indicator for establishing an HSI-based prediction model, where the SNV-PLSR model was checked with R2p of 0.84, RMSEP of 0.38 and RPD of 3.79.

Keywords: Hyperspectral imaging, High-throughput sequencing, Chicken breast meat, Freshness, Bacteria

Absrract

In this article, the hyperspectral imaging technique and the high-throughput sequencing were combined to construct prediction models for the freshness of chicken breast meat. The quality indicators including color, pH, TVC, TVB-N and TBARS were measured to reflect the freshness changes of chicken breast meat under 4 ℃ storage. Meanwhile, spectral images of chicken breast meat were obtained using visible near-infrared (400-1,000 nm) hyperspectral imaging. Through high-throughput sequencing, the major spoilage bacteria including Pseudomonas, Brochothrix and Escherichia were screened out to construct the models for predicting chicken freshness. After spectral preprocessing and characteristic wavelength selection, the prediction models were established using partial least squares regression (PLSR) and support vector machine (SVM). Among the models, the SNV-PLSR model based on characteristic wavelength for Pseudomonas content (Rp2=0.84, RMSEP=0.38, RPD=3.79) posed stronger predictive and generalization abilities. Therefore, the Pseudomonas count was chosen as a characteristic indicator for establishing an HSI-based prediction model to reflect the freshness of chicken breast meat.

Introduction

In 2023, the global chicken production was 102.4 million tons, and China's production was 14.3 million tons, ranking the third among the world (Data from the United States Department of Agriculture). Chilled chicken meat, with its high safety and nutritional value, occupies a large portion of China's chicken meat consumption market (Fang et al., 2023; Zhou et al., 2022). However, chilled chicken meat is easy subject to spoilage by microbial contamination and endogenous enzyme decomposition, accompanied by sticky surfaces and off-flavors. Traditionally, sensory, physicochemical and microbiological properties are widely used methods to check the freshness of chilled meat. The parameters of color, pH, total volatile basic nitrogen (TVB-N) value, thiobarbituric acid reactive substances (TBARS) value, total viable count (TVC) and microbial loads of Pseudomonas and Enterobacteriaceae are common indicators to reveal the freshness of chicken or chicken products (Tian et al., 2022). However, traditional approaches are hard to realize fast and non-destructive detection of chicken meat freshness. Consequently, the exploitation of fast, accurate and non-destructive techniques is significant to improve the efficiency of chicken meat freshness predication.

Hyperspectral imaging (HSI) combines computer engineering and fiber-optic signal processing techniques, which are widely available for acquiring spatial and spectral information of samples through linear or areal scanning (Patel et al., 2024). HSI is faster and more convenient than traditional detection methods, and can realize real-time online detection. In addition, HSI can keep the samples non-destructive and is suitable for the detection of samples in various forms and states (Zhou et al., 2022). The application of HSI includes remote sensing, biomedicine, food quality, agriculture and ecology. In the food sector, HSI is capable of inspecting quality, adulteration and drug residues (Cheng et al., 2024). Currently, the HSI also conducts a deal of in-depth research in the area of meat freshness monitoring. Li et al. (2023) evaluated visual near-infrared spectrum (Vis-NIR) for the freshness prediction of chicken fillets, with an RP2 of 0.93 and an RMSEP of 0.19 for revealing the TVC content and an RP2 of 0.87 and an RMSEP of 2.61 for TVB-N content. However, the routine HSI detection is generally combined with total microbial content to reveal the freshness of meat, which is not differentiated enough to accurately reflect the changes of specific bacteria during storage. Based on this status, the current study proposed that high-throughput sequencing could be combined with HSI to improve the validity and accuracy of meat freshness detection.

In this study, we first reflected the trend of chicken breast meat freshness by measuring the indicators such as TBARS, TVB-N and TVC. Then the high-throughput sequencing was combined with HSI and the representative bacteria in high-throughput sequencing were used as the focus of hyperspectral prediction to construct freshness prediction models. Finally, the prediction models were developed from the representative bacteria, which would realize the high-accuracy detection of chicken breast meat freshness.

Materials and methods

Materials

The Yellow-feathered broilers were purchased from Lihua Livestock & Poultry Co., Ltd (Changzhou, China). Sterile sampling bags were purchased from Biosharp Co., Ltd (Anhui, China). Plate counting agar (PCA), centrimide fucidin-cepha loridine (CFC) and streptomycin sulphate thallous acetate cycloheximide (actidione) agar (STAA) medium and additives were supported by Solarbio Technology Co., Ltd (Beijing, China). The E.Coli/Coliform Chromogenic medium was purchased from Haibo Technology Co., Ltd (Qingdao, China). The E.Z.N.A. ®DNA Kit was purchased from Feiyang Co., Ltd (Guangzhou, China). All other chemicals were purchased from Solarbio Co., Ltd (Beijing, China).

Sample preparation

The chicken breast meat samples were chopped into square pieces of approximately 4 cm*4 cm*2 cm and refrigerated at 4 ℃ in sterile sampling bags. All samples were tested within 7 days (d 0, 1, 2, 3, 4, 5, 6, 7). The 15 samples were taken out per day and kept at room temperature for 30 min, before continuous scanning using HSI. During the storage period, 3 parallel samples were randomly selected daily for measuring freshness indicators and bacterial diversity analysis, and each test was performed in triplicate copies.

Meat quality measurement

Color. The color of samples was measured by the portable colorimeter (CR-13, Konica Minolta, Inc., Tokyo, Japan) under illuminate C, 8 mm diameter aperture and 2°standard observer (Jiang et al., 2024). Before measuring, the colorimeter was adjusted and calibrated with a whiteboard (Y at 85.5, x at 0.3197, y at 0.3374) to ensure the accuracy.

pH. The pH of samples was measured by the portable pH meter (205-PH1, Testo AG, Germany). The electrode was inserted into the chicken breast meat to a depth of 1 cm for measurement (Ye et al., 2021). Prior to measurement, the pH meter was calibrated with standard buffers (pH=4.01, pH=7.01, pH=10.01).

TBARS. The content of TBARS were measured according to the methods of Jiang et al. (2024). Firstly, the 2 g of minced samples were homogenized (4 ℃, 12,100 g, 3 × 20 s) with 10 mL trichloroacetic acid (TCA, 0.75 g) and ethylene diamine tetraacetic acid (EDTA, 0.01 g). The supernatant was collected after centrifuging (4 ℃, 12,100 g, 5 min). Secondly, the 2 mL of thiobarbituric acid solutions (TBA, 0.02 M) were mixed with same content of supernatant at 25 ℃, which was stirred for 30 s, and followed by water bath for 30 min (95 ℃). At last, the supernatant was cooled to 25 ℃ and determined at 532 nm in multimode microplate reader (M2e, Molecular Devices, Inc., Germany). The TBARS values were measured according to the standard curve of malondialdehyde (MDA) and were displayed in mg MDA/kg meat.

TVB-N. The TVB-N content was assayed following with the method of Liu et al. (2020). Firstly, the 3 g of samples and 1 g of magnesium oxide were augmented to the digestive tube coupled with 0.1 M hydrochloric acid as titration solution. The AKjeldahl nitrogen analyzer (Hanon K9860, Jinan, China) of automatic model was applied for quantitative analysis. Finally, the data were reflected as TVB-N mg per 100 grams of meat.

TVC. The content of TVC was calculated and modified according to the method of Wang et al. (2019). Under the sterile condition, the 10 g of samples in the minced state were put in a sterile sampling bag with 90 mL of 0.85 % normal saline, and then shaken for 2 min. Afterwards, the 500 µL of bacterial solution and 4,500 µL of sterile saline were vortex mixed to make a 10-fold sample homogenization. Finally, the 1 mL of diluted sample homogenate was added to the Petri dish and mixed with plate counting agar medium, and TVC values were determined after 48 h of incubation at 37 ℃. The results of testing data were converted as logarithms of colony-forming units (cfu/g).

Bacterial community analysis

The chicken breast meat samples were gently wiped with the cotton head part on a sterile swab for 30 s. After sampling, the swab was placed in a sterile cryovial and preserved at −80 ℃ for testing. The DNA in the swab was extracted with the E.Z.N.A. ®DNA Kit, and the purity and concentration was detected by NanoDrop ND 2000 (Thermo Fisher Scientific, Waltham, MA, USA). Afterwards, the completeness of the genomic DNA was determined through 2 % gel electrophoresis detection. The PCR product electrophoresis had a single wavelength of interest around 500 bp.

The extracted DNA was used as a template for sequencing the V3-V4 region on 16S rRNA based on the second-generation Illumina high-throughput sequencing platform (https://www.illumina.com.cn/systems/sequencing-platforms.html). The sequencing regions were 341F-806R, with primers 341F (5′-CCT AYGGGRBGCASCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′), which were synthesized with specific primers with barcode by Shanghai Lingen Biotech Ltd (Shanghai, China). Finally, the sequencing data were optimized and analyzed using the Biomicroclass cloud platform (http://www.cloud.biomicroclass.com/CloudPlatform) (Yi et al., 2024). Here, the high-quality sequences from individual samples were split according to the barcode and primer sequence. Then, the repeating and singleton sequences were removed and the chimera was filtered according to the positive and negative barcode and primer direction. Finally, the identity criterion was set to 100 % for clustering, and the DADA2 v1.18.0 denoising method was used to obtain amplicon sequence variants (ASV) data.

The alpha diversity indices, comprising Observed species, Chao1, ACE, Shannon, and Simpson were used to reflect the diversity and richness of bacterial communities. The top 10 species were selected to construct the stacked impact map of bacterial community structure at levels of phylum and genus. The abundance of species composition at taxonomic levels of phylum and genus was counted to facilitate the observation of the variation of relative species abundance.

Detection of major spoilage bacteria

The standards of SN/T 4044 (2014), ISO 13722 (1996) and SN/T 0169 (2010) were used for counting Pseudomonas, Brochothrix and Escherichia with slight modification. The sample solution was prepared according to the SN standard. The Pseudomonas and the Brochothrix were cultured on CFC and STAA medium, separately, at 25 ℃ for 48 h. The Escherichia was cultured on E.Coli/Coliform chromogenic medium at 37 ℃ for 24 h. The results of bacterial analysis were reflected as cfu/g.

Acquisition and correction of HSI image data

Spectral information of all chicken breast meat samples was collected using a hyperspectral imaging equipment (Isuzu Optics Co., Taiwan, China). Firstly, the equipment was warmed up for 30 min before scanning. The resolution of the Vis-NIR system (with hyperspectral imaging model) was 2.8 nm along with the wavelength of 400-1,000 nm in reflectance model. In accordance with Zhou et al. (2022), the working parameters were set as below: light source intensity of 45 W, exposure time of 4 ms, distance between camera and sample of 28 cm, and conveyor moving speed of 7.23 nm/s. With the aim of eliminating the adverse effects of external influences on image acquisition, the original image was corrected for black and white. The corrected image was obtained by Formula 1.

Ical=Iraw−IdarkIwhite−Idark (1)

Ical is the corrected image, Iraw is the initial reflection image, Iwhite is the all-white reflection image, and Idark is the all-black reflection image.

Data processing and modeling

In order to avoid background interference, the region of interest (ROI) in the hyperspectral image was extracted through a threshold segmentation algorithm, and the average spectral value of the whole pixels was adopted as the spectrum of each sample. The initial image obtained by HSI was a full wavelength image, and the characteristic wavelength with the least redundant information was selected by employing a successive projection algorithm (SPA). To eliminate noise interference during the acquisition of spectral information, the spectra of the samples required preprocessing. Currently, the preprocessing methods including standard normal variate (SNV), probabilistic quotient normalization (PQN), multiplicative scatter correction (MSC) as well as orthogonal signal correction (OSC) were applied in current study.

The spectral data set was split into a modeling set (n = 90) and a prediction set (30 n = 30) in a 3:1 ratio. The prediction accuracy of support vector machine (SVM) and partial least squares (PLSR) models based on different spectral preprocesses were constructed and compared. Root mean square error (RMSE), coefficient (R2), and also relative percent deviation (RPD) were used as indicators to evaluate the accuracy of prediction models. The R2 shows the coefficient between prediction model and the measured metrics, where the closer the R2 value is to 1, the more acceptable the model is. The RMSE effectively evaluates the stability of the model with the smaller of RMSE indicating the more accurate of this model. Besides, the RPD reflects the robustness of the model, with values of RPD < 2.5 reflecting that the model has a poor predictive ability, values in the range of 2.5 to 3.0 reflecting that the model has a favorable predictive ability, and values of RPD >3.0 reflecting that the model has a strong predictive ability (Zhou et al., 2022).

Statistics analysis

All figures were recorded as mean and standard deviations and statistically analyzed by ANOVA (P < 0.05) with IBM SPSS Statistics 26 software. The heat map of pearson correlation coefficient was calculated and drawn through Origin software (2022).

Results and discussion

Meat qualities

Color and pH changes during storage. Color is the most intuitive indicator of chicken freshness (Sujiwo et al., 2018). As shown in Fig. 1, the lightness (L*) of chicken increased significantly during d 2-4 (P < 0.05), but showed no significant changes among d 6 and d 7 (P > 0.05), the redness (a*) showed no significant changes during d 0-2 and d 3-5 (P > 0.05), and the yellowness (b*) values showed an increasing trend. In the early stage, the dark purple colored myoglobin was easily oxidized to form the bright red colored oxymyoglobin, and in the late stage, the oxymyoglobin was further oxidized to metmyoglobin along with generation of brown color (Fu et al., 2017). Therefore, L* and a* values exhibited an upward and then downward trend, while the b* value showed a slight upward trend, which is consistent with the previous report of Alvarenga et al. (2019). Overall, the color values could not significantly reflect changes of chicken breast meat freshness.

Fig. 1.

Fig 1

Open in a new tab

Trend of freshness indicators of chicken breast meat during storage.

(a) L* (b) a* (c) b*(d) pH (e) TBARS(f) TVB-N (g) TVC.

Note: Different letters (a-e) represent significant differences (P < 0.05, n = 3).

During storage, the lactic acid accumulated in chicken breast meat would decompose into carbon dioxide, alcohol and water, meanwhile the proteins would decompose into amino acids, biogenic amines and other alkaline nitrogen-containing compounds, which result in a gradual rise of pH (Sujiwo et al., 2018). Thus, pH is also an essential metric for meat freshness. According to Kim et al. (2022), the pH value of approximately 6.2-6.5 indicated the commencement of deterioration in chicken breast meat. As revealed in Fig. 1d, the pH range of chicken breast meat was between 5.56 and 5.86, and the results had no significant differences (P > 0.05). Hence the pH values could not be used as remarkable predictors of chicken breast meat freshness.

The indicators of TBARS, TVB-N and TVC. During storage, the fat oxidation is generally appeared and then inspires the decomposition, which finally result in the producing of aldehydes and ketones, along with the increasing of the TBARS content (Park et al., 2023). In Fig. 1e, the TBARS values showed a significant rise during d 3-6 (P < 0.05), and values on d 6 and d 7 were higher than those during d 0-5. When the TBARS value was above 0.5 mg MDA/kg, the rancid odor produced by fat oxidation in the meat could be clearly detected (Zhang et al., 2011). In current study, the TBARS value of chicken breast meat was higher than 0.5 m g/kg after d 4 when a ‘hot odor’ was obviously detected. Therefore, the TBARS can be used as characteristic indicators to reveal the freshness of chilled chicken meat.

TVB-N is the sum of various alkaline substances generated by protein decomposition by the action of specific microorganisms and enzymes, and it represents a significant indicator to assess the meat freshness (Liu et al., 2020). According to China GB/T 5009.1 (2003), the standard of TVB-N in meat is up to 15 mg/100 g for first grade and up to 20 mg/100 g for second grade of fresh meat. In current study, the TVB-N content significantly increased during d 3-5 (P < 0.05), and the values on d 7 were higher than those during d 0-6 (Fig. 1f). The TVB-N content was 15 mg/100 g at d 5 and over 20 mg/100 g at d 7. Thus, the chicken breast meat became sub-fresh at d 5 and spoiled at d 7 in current study, which was comparable to the outcome of Kim et al. (2022).

The proliferation of microorganisms is the main cause of meat deterioration. According to China GB 16869 (2005), the logarithmic value of TVC in fresh and frozen poultry products should be less than 6 log CFU/g. In Fig. 1g, the TVC of the samples increased significantly during storage (P < 0.05), where the original colony count was 3.97 log CFU/g at d 0, and then promoted to 6.37 log CFU/g at d 5. Hence, the chicken breast meat was already spoiled at d 5, which is comparable to the outcome of Tao et al. (2015). In general, TBARS, TVB-N and TVC values could be used as indicators of freshness in combination with HSI to build the prediction models.

Currently, the HSI has been applied for the detection of meat qualities. Tang et al. (2023) developed an artificial neural network (ANN)-based HSI prediction model for color and pH to characterize pork freshness. In the modeling results, the Rp2 of L was 0.9275 and RMSEP was 0.1889, and the Rp2 of pH was 0.8652 and RMSEP was 2.6094, respectively. It showed that the HSI prediction model using color as an indicator performed better in predicting pork freshness. However, based on the current results, the changes of color and pH were not suitable as predictors for chicken breast meat freshness, proving that different raw materials could have different characteristic indicators when combined with HSI. Xiong et al. (2015) established the SPA-PLSR model for TBARS with Rp2 of 0.801 and RMSEP of 0.157. This model validated the potential of TBARS in assessing chicken freshness, but the relatively low Rp2 indicated that the accuracy of the prediction model required further improvement. Li et al. (2023) employed Vis-NIR combined with PLSR to predict the mass indicators of chicken slices, with an Rp2 of 0.9275 and an RMSEP of 0.1889 for TVC, while TVB-N had an Rp2 of 0.8652 and an RMSEP of 2.6094. However, this prediction model (TVC as the indicator) had a RPD value of less than 2.5, which was not accurate enough to applied in different samples. Meanwhile, the HSI prediction model using TVB-N as an indicator had a large RMSEP value, indicating that this model was lacking in stability and accuracy. Based on the correlation with chicken breast meat freshness, the excavation of representative characteristic indicators was necessary to boost the validity of the prediction model.

Bacterial richness and diversity

In incorporation with HSI, the characteristic indicators for reflecting the freshness of chicken breast meat were screened by high-throughput sequencing. After quality control processing, the 453,349 sequences with an average length of 427.86 bp were obtained from 24 sets of samples and grouped as OTUs at 97 % nucleotide identity. In Table S1, the Chao1 and ACE indices were maximal at d 0, while Shannon and Simpson indices reached a minimum value at d 3 and rose at d 7. The magnitude of Chao1 and ACE indices was proportional to the abundance of the community, and the magnitude of Shannon and Simpson indices was proportional to the diversity of bacterial community. Hence, the species content was highest at d 0. In the pre-storage period, the bacteria proliferated to reach a logarithmic growth period, and the competitive inhibition of the major spoilage bacteria led to the decreased colony abundance and bacterial diversity. In the later stage, the competitive effect of the dominant bacteria diminished, resulting in a stable change in abundance and a slight increase in the diversity index (Lei et al., 2022).

Bacterial community composition

The relative abundance (%) of bacterial communities at d 0, 3, 5 and 7 was analyzed for phylum and genus (Fig. 2a). In current study, the top four phylum level included Proteobacteria, Firmicutes, Bacteroidetes and Actinobacteriota. Throughout the storage period, the Proteobacteria was the predominant phyla, with an abundance of more than 80 %. The Firmicutes increased from 6.52 % (d 0) to 21.84 % (d 7), emerging as the second most dominant phylum in the late storage period. According to the results, the phyla Proteobacteria or Firmicutes are the main bacteria implicated in meat spoilage.

Fig. 2.

Fig 2

Open in a new tab

The relative abundance of bacteria at the phylum (a) and genus (b) levels of chicken breast meat during storage (top 10, n = 3).

At the genus level, the bacterial community composition fluctuated significantly throughout storage (Fig. 2b). The top four species at the genus level included Photobacterium, Pseudomonas, Acinetobacter and Brochothrix. There were relatively more species of bacterial flora in the early storage. Among the genera with large changes, the relative abundance of Pseudomonas, Brochothrix and Escherichia showed a gradual increase during storage (Fig. S1). Here, the Pseudomonas was the dominant spoilage genus in the pre-storage period accounting for 16.02 % at d 0, and its abundance increased slowly with storage time, reaching about 25.26 % at d 7. Besides, the Brochothrix and Escherichia were competitively inhibited by dominant species such as Pseudomonas in the pre-storage period, and their abundance was relatively low. The percentage of detected Brochothrix and Escherichia increased significantly at d 7, indicating that the inhibitory effect of the dominant flora had decayed substantially at the late storage period.

Trends in growth of dominant bacteria

Since high-throughput sequencing cannot directly quantify bacterial content, we complemented this approach with traditional methods (plate counting) to determine the absolute quantities of specific bacteria over the 0-7 day period. The contents of Pseudomonas, Brochothrix and Escherichia were measured to further reflect the changes in the freshness of chicken breast meat. Pseudomonas, as the dominant bacteria, was capable of growing and multiplying at a higher rate than Brochothrix and Escherichia during storage. In Fig. 3, the original Pseudomonas count was 2.6 log CFU/g, and exceeded 6 log CFU/g at d 4, which was higher than the national standard limit (Wang et al., 2022). The original Brochothrix count was 2.2 log CFU/g and reached a criterion limit at 6 log CFU/g at d 7. The original Escherichia count was approximately 1.5 log CFU/g and did not exceed the standard limit within 7 d (Zhou et al., 2022). Generally, the counts of Pseudomonas, Brochothrix and Escherichia showed a general increasing trend during storage.

Fig. 3.

Fig 3

Open in a new tab

Variation trend of the contents of three spoilage bacteria during storage.

(a) Pseudomonas (b) Brochothrix (c) Escherichia.

Note: Different letters (a-g) represent significant differences (P < 0.05, n = 3).

Correlation analysis of freshness indicators

During storage, a series of biological and chemical reactions together promoted the change of chicken quality (Xiong et al., 2015; Zhang et al., 2013). In current study, the diverse indicators of chicken breast meat freshness were determined and correlated as illustrated in Fig. 4. Among the quality indicators, the TVB-N was regarded as an objective indicator for revealing the freshness of meat (Jiang et al., 2024). Therefore, the TVB-N value was chosen as the major indicator to analyze the correlation between other parameters. Pseudomonas, Brochothrix and Escherichia were the major spoilage bacteria responsible for the metamorphism of chilled chicken meat (Wang et al., 2022). In existing studies, bacterium such as Pseudomonas spp., Brochothrix, Escherichia and Lactobacillus spp. have been extensively utilized as critical biomarkers in hyperspectral imaging (HSI) models for assessing meat freshness (Jiang et al., 2021; Li et al., 2021; Zhou et al., 2022). The correlation coefficients between TVB-N and TBARS, TVC, Pseudomonas, Brochothrix, and Escherichia were 0.96, 0.97, 0.93, 0.93 and 0.92, respectively (Fig. 4). The microbial indicators showed significant correlations with TVB-N, which effectively reflected the changes in the freshness of the samples, which were appropriate as characteristic indicators for predicting freshness, and thus the microbial content changes were more accurate in constructing the freshness prediction model with HSI.

Fig. 4.

Fig 4

Open in a new tab

Correlation coefficient of freshness indicators during storage of chicken breast meat.

Note: The shade of the circle represents the absolute value of the correlation coefficient, where red represents positive correlation, and blue represents negative correlation. * represents P < 0.05 (n = 3).

Spectral characteristics analysis of HSI

As shown in Fig. 5a, the spectral images of chicken breast meat were acquired from the hyperspectral imaging system during storage. In fresh condition, the large molecules in chicken remained undecomposed and posed to have the highest reflectance. With the extension of storage time, the large molecules were decomposed into small molecules that were easily penetrated, causing a decrease in spectral reflectance (Zhou et al., 2022). Hence, the reflectance spectra showed a decreasing trend as shown in Fig. 5b. In addition, the HSI data obtained from ROIs was collected in the region of 400-1,000 nm. The spectral curves were observed with clear absorption peaks in the region of 430, 550, 560, 700 and 980 nm (Fig. 5c). Among them, the absorption peaks of oxymyoglobin, deoxyhemoglobin and oxyhemoglobin were observed at 430, 550 and 596 nm (Bowker et al., 2014). The absorption peaks in the range of 690-720 nm were strongly associated with water and lipid oxidation (Liu et al., 2022). In special, the absorption peaks of the 3-fold NH3 and hemoglobin were observed around 780 nm (Chen et al., 2018), and the absorption peaks of water were observed around 980 nm (Yuan et al., 2020). Evidently, the spectroscopic data were extensive and contained a huge amount of information, thus the application of spectroscopic data in conjunction with stoichiometric methods was required to develop more accurate predictive models.

Fig. 5.

Fig 5

Open in a new tab

Reflectance spectra of chicken breast meat during storage.

(a) Images of samples collected by HSI during storage. (b) Trends in reflectance spectra during storage. (c) Raw average spectral curves extracted in the region of interest (n = 15).

HSI model development

Compared with the model of TVC, the HSI-based prediction models were developed with the contents of Pseudomonas, Brochothrix and Escherichia as indicators, which were constructed to select the best model for reflecting the chicken freshness. Here, the spectral data were categorized into 90 calibration sets and 30 validation sets, and the collected samples included samples in fresh, sub-fresh, and metamorphic stages, which would enable the model with a wide range of predictions. The spectral reflectance of samples was interfered by external environmental factors, resulting in the extracted spectral information containing noisy information, so the spectral information was preprocessed before the model was established (Jin et al., 2023). In the full-wavelength modeling results for Pseudomonas content, the PLS models were more accurate and robust than the SVM models. The best modeling result was the SNV-PLSR model, with modeling results of Rp2 of 0.76, RMSEP of 0.31 and RPD of 3.47, respectively. Subsequently, the 14 characteristic wavelengths were screened as 382, 385, 386, 387, 390, 391, 425, 476, 547, 574, 600, 631, 698, and 993 nm, respectively. The PLS models also showed a higher accuracy and robustness than the SVM models in the characteristic wavelength modeling results. The optimum prediction model for the characteristic wavelength was the SNV-PLSR model, which was modeled with R p2 of 0.84, RMSEP of 0.38 and RPD of 3.79, respectively (Table 1), with higher Rp2 value, lower RMSEP value and elevated RPD value. Compared to the full-wavelength band, the characteristic wavelength-based HSI prediction model effectively removed redundant wavelengths from the spectral set, thereby improving modeling effectiveness (Zhou et al., 2022). According to the current results, the most effective model with Pseudomonas content was the SNV-PLSR in characteristic wavelength.

Table 1.

Results of Pseudomonas count prediction model for chicken breast meat based on full and characteristic wavelength.

Processing full wavelength
 optimal wavelength
Calibration
 Prediction
 Calibration
 Prediction
Model Pretreatment Rc2 RMSEC Rp2 RMSEP RPD Rc2 RMSEC Rp2 RMSEP RPD
None 0.67 0.26 0.53 0.31 2.33 0.76 0.32 0.66 0.31 2.84
SNV 0.80 0.28 0.76 0.31 3.47 0.84 0.35 0.84 0.38 3.79
PLS MSC 0.69 0.36 0.79 0.32 3.44 0.70 0.37 0.84 0.38 3.48
OSC 0.74 0.30 0.61 0.35 2.22 0.76 0.30 0.65 0.34 2.91
Smoothing 0.65 0.33 0.80 0.34 2.85 0.68 0.27 0.77 0.28 3.08
PQN 0.66 0.28 0.68 0.34 1.81 0.68 0.30 0.70 0.31 2.04
None 0.67 0.28 0.59 0.31 2.32 0.68 0.28 0.61 0.31 2.70
SNV 0.74 0.26 0.71 0.36 3.15 0.80 0.30 0.85 0.37 3.45
SVM MSC 0.74 0.26 0.71 0.36 3.11 0.76 0.36 0.76 0.31 3.21
OSC 0.70 0.40 0.65 0.35 1.96 0.76 0.29 0.67 0.31 2.98
Smoothing 0.67 0.32 0.78 0.31 1.93 0.74 0.33 0.78 0.30 2.12
PQN 0.62 0.34 0.66 0.35 1.80 0.63 0.33 0.74 0.31 2.27
Open in a new tab

Note: Rc2 stands for determination coefficient of calibration, RMSEC stands for root mean square error of calibration, Rp2 stands for determination coefficient of prediction, RMSEP stands for root mean square error of prediction, RPD stands for relative percent deviation, PLS stands for partial least squares, SVM stands for support vector machine, SNV stands for standard normal variate, MSC stands for multiplicative scatter correction, OSC stands for orthogonal signal correction, PQN stands for probabilistic quotient normalization (n = 120, 90 for the modeling set and 30 for the prediction set).

In models constructed with TVC, Brochothrix and Escherichia, characteristic wavelengths also showed a higher accuracy and stability than full-wavelength models. The optimal prediction model for Brochothrix count was OSC-SVM, with modeling resulted in Rp2 of 0.79, RMSEP of 0.32 as well as RPD of 3.84, respectively (Table S2). The best prediction model for Escherichia count was the SNV-PLSR model, with modeling resulted in Rp2 of 0.84, RMSEP of 0.27 and also RPD of 3.74, respectively (Table S3). The best prediction model for TVC was the OSC-PLSR model Rp2 of 0.84, RMSEP of 0.30, along with RPD of 3.07, respectively (Table S4). The RPD values of the best prediction models for TVC, Pseudomonas, Brochothrix and Escherichia were all greater than 2.5, indicating that the models possessed an obvious application potential.

For all the predictive models, the SNV-PLSR model for Pseudomonas possessed higher Rp2 and lower RMSEP, and the RPD value was more than 2.5. In the available studies, Jiang et al. (2021) correlated the total counts of Pseudomonas and Enterobacteriaceae (PEC) in fresh chicken, where the SNV-PLSR model had Rp2 of 0.954, RMSEP of 0.397 and RPD of 3.32, respectively. Compared to previous models, the SNV-PLSR model for Pseudomonas count had lower RMSEP values and higher RPD values, resulting in a better model accuracy and robustness. Pseudomonas spp. demonstrates rapid proliferation and significant spoilage potential. This bacterium efficiently metabolizes nutrients including glucose, free amino acids, and lactic acid, and catalyzes the conversion of inosine or IMP into hypoxanthine, which act as an important metabolic biomarker of chilled chicken spoilage (Zhang et al., 2021). These characteristics establish Pseudomonas count as a reliable indicator for the initiation of spoilage and a characteristic indicator for assessing the freshness of chilled chicken. Hence, Pseudomonas content was chosen as the characteristic indicator combined with HSI to predict chicken freshness. However, the Rp2 value of the HSI prediction model built with Pseudomonas content was less than 0.9, indicating that there remained potential for increasing the accuracy of the model. Deep learning is the procedure of either studying the inherent patterns and representation levels of data, as well as the information obtained from studying procedures which greatly assists in the interpretation of data. As deep learning technology advances swiftly and computational power improves, the deep learning has been applied to the domain of HSI categorization (Jia et al., 2021). In the following study, the deep learning technology could be helpful to select the characteristic index with a view to develop a more accurate and stable prediction model.

Conclusion

In current study, the HSI and characteristic indicators were combined to construct a chicken breast meat freshness prediction model. Firstly, the trends of quality indicators of chicken breast meat were measured during storage (4 ℃). Secondly, the high-throughput sequencing was checked to detect the dominant bacteria in chicken breast meat and the Pseudomonas, Brochothrix and Escherichia were selected as major indicators. There was a strong correlation between the counts of Pseudomonas, Brochothrix, Escherichia and the chicken freshness. Finally, the spectral image information was analyzed based on Vis-NIR HSI. After spectral preprocessing and characteristic wavelength selection, the PLSR and the SVM were applied to establish prediction models. The results of the optimal prediction models were as follows: Pseudomonas (Rp2=0.84, RMSEP=0.38, RPD=3.79), Brochothrix (Rp2=0.79, RMSEP=0.32, RPD=3.84), Escherichia (Rp2=0.84, RMSEP=0.27, RPD=3.74) and TVC (Rp2=0.84, RMSEP=0.30, RPD=3.07). Among the prediction models, the SNV-PLSR within Pseudomonas content had stronger predictive and generalization abilities in characteristic wavelength. Hence, the Pseudomonas content was chosen as the characteristic indicator combined with HSI to predict the chicken freshness. To summarize, the research demonstrates the potential of HSI to predict spoilage bacteria in chilled chicken meat and thus serves as an efficient analytical instrument for monitoring the freshness of meat throughout the storage period.

Declaration of competing interest

We would like to submit the enclosed manuscript entitled “Prediction of chicken breast meat freshness based on hyperspectral imaging technique and high-throughput sequencing”, which we wish to be considered for publication in “Poultry Science”. No conflict of interest exits in the submission of this manuscript, and manuscript is approved by all authors for publication. I would like to declare on behalf of my co-authors that the work described was original research that has not been published previously, and not under consideration for publication elsewhere, in whole or in part. All the authors have approved the manuscript that is enclosed.

Acknowledgements

This work was financially supported by the National Key R&D Program of China (2022YFD2100500).

Footnotes

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.psj.2025.104999.

Appendix. Supplementary materials

mmc1.doc (817.5KB, doc)

References

  1. Alvarenga T.I.R.C., Hopkins D.L., Ramos E.M., Almeida A.K., Geesink G. Ageing-freezing/thaw process affects blooming time and myoglobin forms of lamb meat during retail display. Meat. Sci. 2019;153:19–25. doi: 10.1016/j.meatsci.2019.02.016. [DOI] [PubMed] [Google Scholar]
  2. Bowker B., Hawkins S., Zhuang H. Measurement of water-holding capacity in raw and freeze-dried broiler breast meat with visible and near-infrared spectroscopy. Poult. Sci. 2014;93:1834–1841. doi: 10.3382/ps.2013-03651. [DOI] [PubMed] [Google Scholar]
  3. Chen W., Feng Y.Z., Jia G.F., Zhao H.T. Application of artificial fish swarm algorithm for synchronous selection of wavelengths and spectral pretreatment methods in spectrometric analysis of beef adulteration. Food Anal. Methods. 2018;11:2229–2236. doi: 10.1007/s12161-018-1204-3. [DOI] [Google Scholar]
  4. Cheng J., Sun J., Shi L., Dai C. An effective method fusing electronic nose and fluorescence hyperspectral imaging for the detection of pork freshness. Food Biosci. 2024;59:10388. doi: 10.1016/j.fbio.2024.103880. [DOI] [Google Scholar]
  5. China entry-exit inspection and quarantine industry standards SN/T 0169 (2010): Determination of coliform, fecal coliform and Escherichia coliin food for import and export. https://std.samr.gov.cn/gb/.
  6. China entry-exit inspection and quarantine industry standards SN/T 4044 Enumeration of presumptive Pseudomonas spp. Meat Meat Prod. For Export. 2014 https://std.samr.gov.cn/gb/ [Google Scholar]
  7. China national food safety standard GB/T 5009.1 (2003): Food hygiene test methods physical and chemical part general provisions. https://std.samr.gov.cn/gb/.
  8. China national food safety standard GB 16869 (2005): Fresh and frozen poultry products. https://std.samr.gov.cn/gb/.
  9. Fang X., Ye H., Zhang S., Guo L., Xu Y., Zhang D., Nie Q. Investigation of potential genetic factors for growth traits in yellow-feather broilers using weighted single-step genome-wide association study. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.103034. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Fu Q., Liu R., Zhou G.H., Zhang W.G. Effects of different packaging systems on beef tenderness through protein modifications. J. Food Process. Preserv. 2017;41:e12740. doi: 10.1111/jfpp.12740. [DOI] [Google Scholar]
  11. International Organization for Standardization ISO 13722 (1996): Meat and meat products-enumeration of Brochothrix thermosphacta-colony-count technique. https://www.iso.org.
  12. Jia S., Jiang S., Lin Z., Li N., Xu M., Yu S. A survey: deep learning for hyperspectral image classification with few labeled samples. Neurocomputing. 2021;448:179–204. doi: 10.1016/j.neucom.2021.03.035. [DOI] [Google Scholar]
  13. Jiang F., Zhang J., Zhang R., Zhang W. Effects of ultrasound-assisted vacuum tumbling on the flavor of spiced beef. Food Biosci. 2024;58 doi: 10.1016/j.fbio.2024.103652. [DOI] [Google Scholar]
  14. Jiang S., He H., Ma H., Chen F., Xu B., Liu H., Zhu M., Kang Z., Zhao S. Quick assessment of chicken spoilage based on hyperspectral NIR spectra combined with partial least squares regression. Int. J. Agric. Biol. Eng. 2021;14:243–250. doi: 10.25165/j.ijabe.20211401.5726. [DOI] [Google Scholar]
  15. Jin S., Liu X., Wang J., Pan L., Zhang Y., Zhou G., Tang C. Hyperspectral imaging combined with fluorescence for the prediction of microbial growth in chicken breasts under different packaging conditions. LWT-Food Sci. Technol. 2023;181 doi: 10.1016/j.lwt.2023.114727. [DOI] [Google Scholar]
  16. Kim Y.Y., Park S.J., Kim J.S., Shin H.S. Development of freshness indicator for monitoring chicken breast quality and freshness during storage. Food Sci. Biotechnol. 2022;31:377–385. doi: 10.1007/s10068-022-01034-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Li X., Hu X., Shi J., Qiu B., Xiao J. Visual detection of microbial community during three bacteria mixed fermentation through hyperspectral imaging technology. eFood. 2021:271–278. doi: 10.53365/efood.k/143830. [DOI] [Google Scholar]
  18. Li X., Cai M., Li M., Wei X., Liu Z., Wang J., Jia K., Han Y. Combining Vis-NIR and NIR hyperspectral imaging techniques with a data fusion strategy for the rapid qualitative evaluation of multiple qualities in chicken. Food Control. 2023;145 doi: 10.1016/j.foodcont.2022.109416. [DOI] [Google Scholar]
  19. Liu H., Ji Z., Liu X., Shi C., Yang X. Non-destructive determination of chemical and microbial spoilage indicators of beef for freshness evaluation using front-face synchronous fluorescence spectroscopy. Food Chem. 2020;321 doi: 10.1016/j.foodchem.2020.126628. [DOI] [PubMed] [Google Scholar]
  20. Liu Y., Long Y., Liu H., Lan Y., Long T., Kuang R., Wang Y., Zhao J. Polysaccharide prediction in Ganoderma lucidum fruiting body by hyperspectral imaging. Food Chemistry-X. 2022;13 doi: 10.1016/j.fochx.2021.100199. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Park S., Yang M., Yim D.G., Jo C., Kim G. VIS/NIR hyperspectral imaging with artificial neural networks to evaluate the content of thiobarbituric acid reactive substances in beef muscle. J. Food Eng. 2023;350 doi: 10.1016/j.jfoodeng.2023.111500. [DOI] [Google Scholar]
  22. Patel D., Bhise S., Kapdi S.S., Bhatt T. Non-destructive hyperspectral imaging technology to assess the quality and safety of food: a review. FPPN. 2024;6:69. doi: 10.1186/s43014-024-00246-4. [DOI] [Google Scholar]
  23. Sujiwo J., Kim D., Jang A. Relation among quality traits of chicken breast meat during cold storage: correlations between freshness traits and torrymeter values. Poult. Sci. 2018;97:2887–2894. doi: 10.3382/ps/pey138. [DOI] [PubMed] [Google Scholar]
  24. Tang X., Rao L., Xie L., Yan M., Chen Z., Liu S., Chen L., Xiao S., Ding N., Zhang Z., Huang L. Quantification and visualization of meat quality traits in pork using hyperspectral imaging. Meat. Sci. 2023;196 doi: 10.1016/j.meatsci.2022.109052. [DOI] [PubMed] [Google Scholar]
  25. Tao F., Peng Y., Gomes C.L., Chao K., Qin J. A comparative study for improving prediction of total viable count in beef based on hyperspectral scattering characteristics. J. Food Eng. 2015;162:38–47. doi: 10.1016/j.jfoodeng.2015.04.008. [DOI] [Google Scholar]
  26. Tian T., Kang Y., Liu L., Wang X. The effect of super-chilled preservation on shelf life and quality of beef during storage. Food Sci. Technol. 2022;42:e73222. doi: 10.1590/fst.73222. [DOI] [Google Scholar]
  27. Wang H., He H., Ma H., Chen F., Kang Z., Zhu M., Zhengrong W., Shengming Z., Zhu R. LW-NIR hyperspectral imaging for rapid prediction of TVC in chicken flesh. Nt. J. Agric. Biol. Eng. 2019;12:180–186. doi: 10.25165/j.ijabe.20191203.4444. [DOI] [Google Scholar]
  28. Wang X., Wang Z., Sun Z., Wang D., Liu F., Lin L. In vitro and in situ characterization of psychrotrophic spoilage bacteria recovered from chilled chicken. Foods. 2022;12:95. doi: 10.3390/foods12010095. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Xiong Z., Sun D.W., Pu H., Xie A., Han Z., Luo M. Non-destructive prediction of thiobarbituricacid reactive substances (TBARS) value for freshness evaluation of chicken meat using hyperspectral imaging. Food Chem. 2015;179:175–181. doi: 10.1016/j.foodchem.2015.01.116. [DOI] [PubMed] [Google Scholar]
  30. Ye Z., Zhang J., Lorenzo J.M., Zhang M., Zhang W. Effects of bromelain on the quality of smoked salted duck. Food Sci. Nutr. 2021;9:4473–4483. doi: 10.1002/fsn3.2422. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Yi Z., Qiu M., Xiao X., Ma J., Yang H., Wang W. Quantitative characterization and dynamics of bacterial communities in ready-to-eat chicken using high-throughput sequencing combined with internal standard-based absolute quantification. Food Microbiol. 2024;118 doi: 10.1016/j.fm.2023.104419. [DOI] [PubMed] [Google Scholar]
  32. Yuan R., Liu G., He J., Ma C., Cheng L., Fan N., Ban J., Li Y., Sun Y. Determination of metmyoglobin in cooked tan mutton using Vis/NIR hyperspectral imaging system. J. Food Sci. 2020;85:1403–1410. doi: 10.1111/1750-3841.15137. [DOI] [PubMed] [Google Scholar]
  33. Zhang W., Marwan A.H., Samaraweera H., Lee E.J., Ahn D.U. Breast meat quality of broiler chickens can be affected by managing the level of nitric oxide. Poult. Sci. 2013;92:3044–3049. doi: 10.3382/ps.2013-03313. [DOI] [PubMed] [Google Scholar]
  34. Zhang T., Ding H., Chen L., Zhang S., Wu P., Xie K., Pan Z., Zhang G., Dai G., Wu H., Wang J. Characterization of chilled chicken spoilage using an integrated microbiome and metabolomics analysis. Food Res. Int. 2021;144 doi: 10.1016/j.foodres.2021.110328. [DOI] [PubMed] [Google Scholar]
  35. Zhou B., Fan X., Song J., Wu J., Pan L., Tu K., Peng J., Dong Q., Xu J., Wu J. Growth simulation of Pseudomonas fluorescens in pork using hyperspectral imaging. Meat. Sci. 2022;188 doi: 10.1016/j.meatsci.2022.108767. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

mmc1.doc (817.5KB, doc)

Articles from Poultry Science are provided here courtesy of Elsevier

RESOURCES