Search

• A dataset of X-ray radiographs of pear fruit was simulated from CT scans. • Deep anomaly detection models were trained on data of healthy pears and tested on data of healthy and defect pears. • Synthetic anomalies in training improved performance over the unsupervised case. • High AUC scores (up to 0.96) and classification accuracy (up to 95%) were obtained. • Defect pears with a cavity percentage > 1.0% could be detected 100% accurate. To preserve the quality of fresh fruit after harvest and to meet the year-round demand for high-quality fruit, pears are stored under a controlled atmosphere. However, due to preharvest events or suboptimal storage conditions, internal disorders might develop resulting in severe quality loss. Examples include internal browning and cavities, which are invisible externally. Here, X-ray radiography is investigated as a technique for internal quality inspection. The detection of defect fruit is approached as an anomaly detection (AD) problem, in which a model is constructed using nominal data and an anomaly score is used to identify defect fruit. In this work, multiple deep AD methods are shown to be effective to detect pears with internal cavity and browning disorders using X-ray radiographs (mean area under the receiver operating characteristic curve (AUC) up to 0.962). The best performing methods were found on par with a state-of-the-art multisensor disorder detection method (mean AUC up to 0.966). By investigating AD performance in function of internal disorder severity, it was shown that defect fruit with a cavity volume percentage > 1.0% could be detected 100% accurate using inline X-ray imaging. For lower cavity area percentages, the accuracy depended on the internal browning severity. Additionally, the explainability of the deep AD methods, i.e., how well human interpretable insight can be provided from each method's predictions, were qualitatively evaluated using anomaly heatmaps, which provided useful insight in the execution of the deep learning algorithms. [ABSTRACT FROM AUTHOR]

• Healthy and defect Conference pears resulted from storage experiments. • X-ray CT scans were taken and internal disorders were manually annotated. • A deep neural network was trained to segment internal disorders automatically. • Quantitative data of the internal quality was used to accurately sort the fruit. • The method can be used for non-destructive internal food quality inspection. Long term storage is required to deliver high quality pear fruit year-round. Under suboptimal storage conditions, internal disorders, such as internal browning and cavity formation, can develop and are often invisible from the outside. We present a non-destructive inspection method to quantify internal disorders in X-ray CT scans of pear fruit using a deep neural network for semantic segmentation. Herein, a U-net based model was trained to automatically indicate healthy tissue, core and regions affected by internal disorders, i.e., cavity formation and internal browning. The quantitative data resulting from the segmentations was used to measure the severity of internal disorders. Excellent classification accuracies of 99.4 and 92.2% were obtained for the classification of "consumable" vs " non-consumable" fruit on the one hand and "healthy" vs "defect but consumable" vs "non-consumable" fruit on the other hand. The identification of "defect but consumable" fruit showed to be the most difficult. [ABSTRACT FROM AUTHOR]

To preserve the quality of fresh pear fruit after harvest and deliver quality fruit year-round a controlled supply chain and long-term storage are applied. During storage, however, internal disorders can develop due to suboptimal storage conditions that may not cause externally visible symptoms. This makes them impossible to be detected by current commercial quality grading systems in a reliable and non-destructive way. A combination of a Support Vector Machine coupled with a feature extraction algorithm and X-ray Computed Tomography is proposed to successfully detect internal disorders in 'Conference' and 'Cepuna' pear fruit nondestructively. Classifiers were able to distinguish defective from sound fruit with classification accuracies ranging between 90.2 and 95.1% depending on the cultivar and number of used features. Moreover, low false positive and negative rates were obtained, respectively ranging between 0.0 and 6.7%, and 5.7 and 13.3%. Classifiers trained on 'Conference' data were transferred effectively to the 'Cepuna' cultivar, suggesting generalizability to other cultivars as well. With continuing developments in both hardware and software to increase inspection speed and reduce equipment costs, the method can be implemented in industrial applications, e.g., inline translational X-ray CT. • Healthy and defect Conference and Cepuna pears resulted from storage experiments. • Fast X-ray CT scans were obtained of all fruit and relevant features were extracted. • SVMs achieved high accuracy in detecting defect fruit regardless of the cultivar. • Feature selection could limit the used features to two without a drop in performance. • Method can be implemented for postharvest fruit sorting using fast X-ray CT. [ABSTRACT FROM AUTHOR]

[Display omitted] • Deep learning was able to label defects in X-ray radiographs of pear fruit. • A large amount of ground truth data were obtained by simulations from X-ray CT data. • Synthetic training data improved segmentation of browning and cavity defects. A particular challenge for food quality inspection remains the quantification of internal defects. This work addresses the challenge of internal defects segmentation in pear fruit (cv. Conference) based on high-throughput inline X-ray radiography images in combination with deep learning. Unlike previous approaches, which focused on healthy vs defect classification, the current segmentation approach is able to determine the type of defect, its dimensions, and location. To this end, a novel simulation method was designed to obtain input-target image pairs of radiography data, and more diverse defect pears were generated using a conditioned generator model. The obtained data contributed to the design of a segmentation model that labeled every pixel in the X-ray radiographs of pear fruit as 'external air', 'healthy tissue', 'core', 'browning', or 'cavity'. We demonstrated that additional synthetic data in the learning process drastically improved the model performance. For instance, the mean IoU increased from 0.781 ± 0.112 to 0.883 ± 0.088 for consumable pears with minor cavities and browning. In terms of utility, our segmentation maps provide more detailed information about the type, size, and location of the disorders compared to the heatmaps produced by existing benchmark classifiers. [ABSTRACT FROM AUTHOR]