Intelligent evaluation of total volatile basic nitrogen (TVB-N) content in chicken meat by an improved multiple level data fusion model

• Published in: Sensors and actuators. B, Chemical Vol. 238; pp. 337 - 345
• Main Authors: Khulal, Urmila, Zhao, Jiewen, Hu, Weiwei, Chen, Quansheng
• Format: Journal Article
• Language: English
• Published: Lausanne Elsevier B.V 01.01.2017; Elsevier Science Ltd
• Subjects: Ant colony optimization (ACO); Aroma; Chickens; Colorimetric sensors array; Colorimetry; Correlation analysis; Data fusion; Data integration; Data processing; Freshness; Hyperspectral imaging (HSI); Improved MLF; Mathematical models; Meat; Multiple level data fusion (MLF); Multisensor fusion; Nitrogen; Odors; Optical measuring instruments; Optimization; Poultry; Poultry meat; Sensors; Texture; LLA; Colorimetric sensors array; VOCs; PCs; PCA-BPANN; RMSEP; Hyperspectral imaging (HSI); TVB-N; MLF; RPD; Multiple level data fusion (MLF); ACO; ROI; RMSECV; HLA; Improved MLF; Poultry meat; Ant colony optimization (ACO); BP-ANN; Rp; HSI; ILA
• ISSN: 0925-4005; 1873-3077
• DOI: 10.1016/j.snb.2016.07.074

•The multiple sensors data fusion model can be used for assessing chicken meat's freshness.•Multiple level data fusion model further improved by correlation analysis.•The sensors fusion yielded better results than either of the individual sensor system. The objective of this paper is to present a fusion model of an odor sensor and highly advanced optical sensor to evaluate total volatile basic nitrogen (TVB-N) content in chicken meat. Here, the aroma or the odor data variables obtained from the odor sensor i.e. colorimetric sensor and the spectral as well as textural data variables obtained from the optical sensor i.e. HSI, were fused together for further data processing. 36 odor variables obtained via the low-level data abstraction (LLA) were simply concatenated with the 30 texture feature variables obtained by middle/intermediate level data abstraction (ILA) totaling to a 66 variables’ dataset. This approach of multiple level data fusion (MLF) produced the better PCA-BPANN prediction results than either of the individual system did, with the higher Rp of 0.8659, lower RMSEP of 4.587mg/100g along with the increased calibration model efficacy. Furthermore, the prediction level escalated with Rp of 0.8819 and RMSEP of 4.3137mg/100g when the data fusion technique was improved by applying Pearson’s correlation analysis and uncorrelated data variables were removed from each of the dataset at the statistical level of significance. This step reduced the data variables but not the original information. Therefore, the results highly encourage multiple sensor fusion and the improved MLF technique for better model performance to evaluate chicken meat’s freshness.