Search papers, labs, and topics across Lattice.
This study benchmarks automated binary classification of hazelnut quality using X-ray images, addressing challenges such as subtle radiographic differences and class imbalance. By evaluating seven single-model configurations and ten probability-aggregation ensembles on a dataset of 799 images, the authors found that an ensemble combining a convolutional neural network and a frozen Swin Transformer achieved the highest mean balanced accuracy of 86.3%. The research underscores the critical role of expert reassessment in improving model performance and highlights the necessity for multi-split evaluation to ensure reliable comparisons in small datasets.
Expert reassessment of ambiguous samples boosted model performance by up to 8.1 percentage points, revealing the hidden potential in small, imbalanced datasets.
Non-destructive X-ray imaging can reveal internal hazelnut defects that are difficult to detect by external inspection alone; however, automated interpretation remains challenging because of subtle radiographic differences among classes, marked class imbalance, and limited annotated data. Here, we present a benchmark for binary hazelnut quality classification (healthy versus defective) based on 799 segmented single-kernel X-ray images (224 x 224 pixels, grayscale), grouped into 101 acquisition units. Seven single-model configurations and ten probability-aggregation ensembles were evaluated using a group-wise split-rotation protocol across five data splits generated using different random seeds. Decision thresholds were selected on the validation set, and performance was assessed deterministically on validation and test sets. Under the expert-reassessed annotation condition, the average-probability ensemble of the binary cross-entropy-trained convolutional neural network and frozen Swin Transformer achieved the highest mean balanced accuracy (86.3% +/- 1.8%, five seeds), with several other ensembles providing comparable performance. Across methods, substantial split-to-split variability was observed, indicating that multi-split evaluation is essential for reliable model comparison at this dataset scale. Expert reassessment of ambiguous samples improved the performance of all 17 evaluated methods by 2.8-8.1 percentage points, while having only a limited effect on cross-split variance. The results highlight both the potential of deep learning for automated X-ray-based hazelnut quality assessment and the importance of rigorous evaluation and label curation in small, imbalanced agricultural imaging datasets.
