Abstract:Spectroscopy-based techniques have made remarkable advancements in their application to fruit quality control but encounter challenges of high model uncertainty arising from biological variability. Minor changes in spectral measurement orientations or positions significantly altered the model prediction for fruit quality, which severely affects the reliability of online fruit grading systems. This study assesses the influence of varying structural designs in deep learning models and develops a Knowledge-Guided Convolutional Neural Network (KGCNN) to mitigate predictive uncertainty caused by different orientations. A novel loss criterion that quantifies model uncertainty incorporating a grouped input strategy is introduced, enabling the implementation of a knowledgeguided approach. Experiments are conducted on Mandarin and Valencia orange datasets collected on an onsite grading system for soluble solid content assessment. The proposed KGCNN model effectively reduces 27.1 % and 25.5 % averaged prediction variance compared with CNN and PLS models on five detections of random orientations, demonstrating a significant mitigation of model uncertainty. This strategy operates within the framework of the developed model without the need for structural hyperparameter adjustments, with a minor impact on the test accuracy, evidenced by a 1.7 % increase in RMSE and a 0.2 % decrease in R2 averaged on two datasets. A hidden feature visualization technique is further employed to explain the working mechanism of the decreased model uncertainty by KGCNN. The visual evidence supports the effectiveness of the knowledge-guided approach in mitigating the nonlinear scatter within the distributions of learned features from input data collected at multiple orientations, attributed to the constraint effect of the introduced loss criterion. This research has the potential to alleviate the stringent requirements on fruit loading or measurement positions within onsite grading systems and enhance the reliability of deep learning models in practical applications.

Predicting the Quality of Tangerines Using the GCNN-LSTM-AT Network Based on Vis-NIR Spectroscopy

Improving SSC Detection Accuracy of Cherry Tomatoes by Feature Synergy and Complementary Spectral Bands Combination

Enhancing model robustness through different optimization methods and 1-D CNN to eliminate the variations in size and detection position for apple SSC determination

Researches of fruit quality prediction model based on near infrared spectrum

Enhancing Spectroscopy-Based Fruit Quality Control: A Knowledge-Guided Machine Learning Approach to Reduce Model Uncertainty

E-nose Combined with Chemometrics to Trace Tomato-Juice Quality

Evaluation of peach quality indices using an electronic nose by MLR, QPST and BP network

Nondestructive Determining the Soluble Solids Content of Citrus Using Near Infrared Transmittance Technology Combined with the Variable Selection Algorithm

Nondestructive Detection of Tomato Quality Based on Multiregion Combination Model

Principal Components-Artificial Neural Networks for Predicting SSC and Firmness of Fruits Based on Near Infrared Spectroscopy

Combining Vis-NIR and NIR Spectral Imaging Techniques with Data Fusion for Rapid and Nondestructive Multi-Quality Detection of Cherry Tomatoes

Integrated Fruit Ripeness Assessment System Based on an Artificial Olfactory Sensor and Deep Learning

Rapid Detecting SSC and TAC of Peaches Based on NIR Spectroscopy

Using Knowledge-Guided Temperature Correction for Online Non-Destructive Detection of Soluble Solids Content in Pear Via Vis/NIR Spectroscopy

Integrating deep learning and data fusion for enhanced oranges soluble solids content prediction using machine vision and Vis/NIR spectroscopy

Advanced Chemometrics Toward Robust Spectral Analysis for Fruit Quality Evaluation

Prediction of loquat soluble solids and titratable acid content using fruit mineral elements by artificial neural network and multiple linear regression

Quality Assessment and Ripeness Prediction of Table Grapes Using Visible–Near-Infrared Spectroscopy

Non-destructive detection of the quality attributes of fruits by visible-near infrared spectroscopy

Fruit Detection Research Based on Near-Infrared Spectroscopy and Lightweight Neural Network

Development of deep learning method for predicting firmness and soluble solid content of postharvest Korla fragrant pear using Vis/NIR hyperspectral reflectance imaging