Akademik Veri Yönetim Sistemi
15th International Congress of the Innovative Agricultural Technologies, IAT 2025, Antalya, Türkiye, 15 - 19 Ekim 2025, cilt.805 LNCE, ss.222-236, (Tam Metin Bildiri)
This study proposes a solution to the problem of fast, non-destructive, and reliable measurement of moisture, a critical factor in grains, utilizing a microwave sensor based on a microstrip ring resonator operating at 2.45 GHz, along with a machine learning-based estimation approach. The objective is to achieve high-accuracy moisture estimation by modeling the relationship between dielectric parameters, such as phase shift, attenuation, dielectric constant (ε′), loss factor (ε″), and loss tangent (tan δ), and rice moisture content. In the methodology, the k-Nearest Neighbor (kNN), Random Forest (RF), and artificial neural network-multi-layer perceptron (ANN-MLP) algorithms were evaluated using a 10-fold cross-validation approach. The greatest R values were obtained from kNN and RF, with the values of 0.978 and 0.977, respectively. The lowest RMSE and MAE values were also obtained in kNN, at 0.660 and 0.480, respectively. The highest RMSE, MAE, RRSE, and RAE values were found in ANN-MLP, with the values of 0.966, 0.751, 30.451%, and 29.056%, respectively. Thus, kNN is comparable to RF, which substantially exceeds ANN-MLP; both kNN and RF attained an error reduction of roughly 79–81% relative to the naive approach. The kNN, RF, and ANN-MLP models chosen for this study have exhibited remarkable precision in predicting moisture content. The integration of these models with IoT and cloud technologies is expected to yield similarly excellent results.