طراحي و توسعه‌ي سامانه‌ي فضاي آزاد براي برآورد غيرمخرب كيفيت داخلي و باروري تخم‌مرغ در محدوده‌ي مايكروويو

The poultry industry is currently one of the largest sectors providing food worldwide. Many efforts have been made to increase its efficiency. One of the key challenges in this industry is determining egg quality, which is influenced by factors such as storage time, storage an‎d transportation conditions, an‎d hen nutrition. Additionally, detecting fertile eggs before incubation is critical for improving poultry production efficiency an‎d enhancing overall industry productivity. Today, the modern poultry industry seeks innovative management methods to replace traditional egg grading an‎d fertility detection approaches, such as optical an‎d visual inspections. In this dissertation, a free-space dielectric system was employed for the non-destructive eva‎luation of the internal quality an‎d fertility of eggs. The measurement system was first simulated using HFSS software. It consists of a network analyzer capable of recording scattering parameters across a specified frequency range (8–12 GHz) an‎d two X-ban‎d coaxial-to-waveguide adapters serving as antennas. In the quality assessment section, the spectra of 252 eggs stored for 1, 3, 5, 7, 15, an‎d 24 days at room temperature were measured in three different orientations (horizontal 1, horizontal 2, an‎d vertical). Reference experiments were conducted immediately after spectral acquisition to determine egg quality indices. Partial least squares (PLS) regression was used to predict the air cell height (ACH), yolk coefficient (YC), thick albumen height (TAH), Haugh unit (HU), an‎d albumen pH. According to the PLS regression results, ACH showed the highest predictive ability with a residual prediction deviation (RPD) of 4.80. YC, albumen pH, TAH, an‎d HU also achieved excellent RPDs of 4.00, 3.96, 3.44, an‎d 3.19, respectively. By examining the effects of spectrum type an‎d egg placement orientation on the RPDs obtained from the best PLS models, the PR_S22 spectrum combined with the horizontal 1 orientation yielded the best performance. Using this configuration, feature selec‎tion methods, including partial least squares-variable importance in projection (PLS-VIP), correlation-based feature selec‎tion (CFS), an‎d competitive adaptive reweighted sampling (CARS), were applied to identify the most influential frequencies. Predictive models based on these frequencies were then developed using an artificial neural network (ANN). Among the feature selec‎tion methods, CARS outperformed the others, producing robust ANN models with excellent RPDs of 4.80 (ACH), 4.00 (YC), 3.27 (TAH), 3.03 (HU), an‎d 3.72 (albumen pH). The soft independent modeling of class analogy (SIMCA) classifier was also used to categorize eggs based on their storage time. SIMCA successfully classified storage times across all three orientations using the RL_S22 spectrum, achieving 100% sensitivity, specificity, precision, accuracy, an‎d F1-score. In the fertility assessment section, the spectra of 100 fertile an‎d 125 infertile eggs were measured on day 0 in two orientations (horizontal 1 an‎d vertical). The eggs were incubated immediately, an‎d additional spectral measurements were taken on days 1, 2, 3, an‎d 4 of incubation. Fertility was confirmed by breaking the eggs on the sixth day of incubation. Several classifiers, including SIMCA, support vector machine (SVM), partial least squares discriminant analysis (PLS-DA), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), an‎d ANN, were eva‎luated for classifying fertile an‎d infertile eggs based on spectra collected before incubation. Across all classifiers, the best results were generally obtained using the IL_S21 spectrum with second derivative preprocessing in the vertical orientation, except for PLS-DA, which performed best with the horizontal 1 orientation.

احمد ميره اي , غلامرضا عسكري

محمد صدقي , عباس همت

مرتضي صادقي , اميرحسين مهدوي , مهدي قاسمي ورنامخواستي