Ripening Degree Assessment of Organic Strawberry Based on Electrical Impedance Spectroscopy With Machine Learning

Assessing the ripening degree of organic strawberries in-situ is crucial for ensuring optimal fruit quality and marketability. However, traditional methods are time-consuming, subjective, and often destructive, as they require fruits to be harvested for evaluation. This study developed a non-invasive, real-time approach for assessing the ripening degree of organic strawberries in-situ using a bioelectrical analysis system and machine learning. By focusing on in-situ measurements, this study distinguished itself from previous studies that primarily focused on harvested fruits. Measurements were collected in-situ from 98 Honeoye strawberries (45 ripe, 53 overripe) at an open organic farm, across a frequency range of 1 Hz to 1 MHz. Cole-Cole plots, Nyquist plots, t-SNE dimensionality reduction, and statistical measures were employed to extract optimal input features from the in-situ data. Feed-forward Neural Network (FFNN), Spiking Neural Network (SNN), and Deep Recurrent Neural Network (DRNN) models were developed, evaluated using stratified 5-fold cross-validation. Hyperparameter tuning was performed using grid search to optimize training parameters. The DRNN model, leveraging long short-term memory (LSTM) layers, achieved the highest performance with 95.2% accuracy, 0.99 AUC-ROC, and 0.98 AUPRC, outperforming all the existing methods. These results demonstrate the effectiveness of the proposed method for objectively classifying strawberry ripeness stages in-situ. Hence, the developed method addresses the limitations of traditional assessments and offers a promising tool for precision agriculture and harvest management, potentially improving fruit quality, reducing waste, and enhancing overall agricultural efficiency by enabling timely and accurate in-situ ripeness assessment.