Applying convolutional neural networks to assess the external quality of strawberries

Abstract

Deterioration in the appearance of strawberries is attributed to rapid metabolic changes, cellular damage, and softening occurring during their distribution and storage. To quickly and non-destructively monitor the external quality of strawberries, recognition models based on 750 Red Green Blue (RGB) image classifications and using convolutional neural networks (CNNs) were developed. A model using eight configurations was used to compare the discrimination accuracies according to the following variables: training–test set distributions, number of learning images, and number of epochs. Strawberry samples were classified as fresh, bruised, or moldy. According to our validation results, training the model with 90 % of the image data ensured a high learning performance. Using our test dataset, we found that the accuracy, precision, specificity, and sensitivity of the model reached 97 %. In the feature map derived from convolutional layers, the bruised and moldy areas of the strawberry were also identified. Together, these results suggest that CNNs have potential use in the non-destructive monitoring of quality changes in the food industry.