Akademik Veri Yönetim Sistemi
15th International Congress of the Innovative Agricultural Technologies, IAT 2025, Antalya, Türkiye, 15 - 19 Ekim 2025, cilt.805 LNCE, ss.195-212, (Tam Metin Bildiri)
The soluble solids content (SSC) of sweet cherries plays a critical role in determining taste, aroma, and flavor attributes that shape consumer preferences, and it serves as a fundamental parameter for farmers and wholesalers in establishing optimal harvest timing. This study demonstrates the feasibility of predicting SSC through a non-invasive approach by integrating data obtained from dielectric spectroscopy with the Convolutional Neural Network (CNN) model, a deep learning technique. Dielectric properties, including the dielectric constant (ε′), loss factor (ε′′), and loss tangent (tan δ), along with SSC values, were measured across the 15–300 MHz range using an open-ended coaxial probe during different harvest periods. The developed CNN model, optimized using Bayesian optimization and early stopping, successfully extracted spatial features from dielectric spectral data, exhibiting superior predictive performance. The model achieved high R2 and correlation values: for the E1 dataset (incorporating ε′), test R2 of 0.7862 and test correlation of 0.8975; for the E2 dataset (incorporating ε′′), test R2 of 0.5668 and test correlation of 0.9131; and for the TANLOSS dataset (incorporating tan δ), test R2 of 0.9099 and test correlation of 0.9566. These results are substantiated by low test MSE (0.6980) and MAE (0.6599) values in the TANLOSS dataset, confirming the model’s predictive accuracy and generalization capability. The findings indicate that reduced frequency ranges preserve accuracy, and a dielectric measurement system operating with fewer frequencies can enable low-cost, high-accuracy SSC prediction.