University of Tehran Press Iranian Journal of Biosystem Engineering 2008-4803 56 2 2025 08 23 Development and Evaluation of an Intelligent Machine Vision System for Detecting Chicken Gizzard Adulteration in Minced Red Meat Development and Evaluation of an Intelligent Machine Vision System for Detecting Chicken Gizzard Adulteration in Minced Red Meat 90 105 104065 10.22059/ijbse.2025.398182.665600 FA Mobin Rezazadeh Mechanical Engineering of Biosystems Department, Faculty of Agriculture, Shahrekord University, Shahrekord, I.R.Iran 0009-0008-7102-6060 Sajad Kiani Biosystems Engineering Department,, Sari Agricultural Sciences and Natural Resources University, Sari, Iran Mahdi Ghasemi Varnamkhasti Mechanical Engineering of Biosystems Department, Faculty of Agriculture, Shahrekord University, Shahrekord, Zahra Izadi Mechanical Engineering of Biosystems Department, Faculty of Agriculture, Shahrekord University, Shahrekord, I.R. Iran. Journal Article 2025 07 08 In this study, an intelligent machine-vision system was developed and evaluated for detecting chicken gizzard adulteration in ground red mutton-veal meat, using digital images captured by a mobile phone. To this end, standard samples of minced red meat (55% mutton and 45% beef) with varying proportions of chicken gizzard (0 to 100%) were prepared, and images were captured in a laboratory environment both directly from the sample surface and through plastic wrap packaging. The color features (RGB) of the images were extracted using the MATLAB image processing toolbox, and modeling was performed employing statistical and machine learning methods, including Principal Component Analysis (PCA), Partial Least Squares Regression (PLSR), and Multilayer Perceptron (MLP) neural networks. The best performance of the linear PLSR model for estimating the adulteration percentage yielded an R²V=0.7 and RMSEV=16.51 under conditions without plastic wrap, whereas the nonlinear MLP model achieved an R²V=0.97 and RMSEV=6.673. These results were lower (0.9 and 12.54) for the data acquisition through plastic wrap due to the light reflections caused by the covering. Furthermore, the MLP classifier achieved classification accuracies of 85%, 96.4%, 92.6%, 73.7%, 76.2%, and 96.7% for adulteration levels 0-10%, 10–20%, 20–30%, 30–40%, 40–50%, and above 50%, respectively. The average precision, sensitivity, and F1 score for the developed model were obtained as 0.975, 0.974, and 0.975, respectively. The results showed that this non-destructive method is both fast and reliable for identifying gizzard fraud in ground meat, and can serve as a basis for developing meat quality control systems. In this study, an intelligent machine-vision system was developed and evaluated for detecting chicken gizzard adulteration in ground red mutton-veal meat, using digital images captured by a mobile phone. To this end, standard samples of minced red meat (55% mutton and 45% beef) with varying proportions of chicken gizzard (0 to 100%) were prepared, and images were captured in a laboratory environment both directly from the sample surface and through plastic wrap packaging. The color features (RGB) of the images were extracted using the MATLAB image processing toolbox, and modeling was performed employing statistical and machine learning methods, including Principal Component Analysis (PCA), Partial Least Squares Regression (PLSR), and Multilayer Perceptron (MLP) neural networks. The best performance of the linear PLSR model for estimating the adulteration percentage yielded an R²V=0.7 and RMSEV=16.51 under conditions without plastic wrap, whereas the nonlinear MLP model achieved an R²V=0.97 and RMSEV=6.673. These results were lower (0.9 and 12.54) for the data acquisition through plastic wrap due to the light reflections caused by the covering. Furthermore, the MLP classifier achieved classification accuracies of 85%, 96.4%, 92.6%, 73.7%, 76.2%, and 96.7% for adulteration levels 0-10%, 10–20%, 20–30%, 30–40%, 40–50%, and above 50%, respectively. The average precision, sensitivity, and F1 score for the developed model were obtained as 0.975, 0.974, and 0.975, respectively. The results showed that this non-destructive method is both fast and reliable for identifying gizzard fraud in ground meat, and can serve as a basis for developing meat quality control systems. https://ijbse.ut.ac.ir/article_104065_4cc27673cfc99671a54483f3d6d36cb9.pdf