Automated Detection of Mechanical Damage in Flaxseeds Using Radiographic Imaging and Machine Learning

The growing demand for flaxseed as a source of healthy edible oil mandates the need for adopting novel strategies for preserving its quantity and quality. Mechanical damage during harvest and handling is one of the important threats that can adversely affect the quality and viability of flaxseeds. Currently, mechanical damage assessment in grains is mainly performed by human visual inspection, which is a subjective and time-consuming procedure. In this study, the authors propose to utilize radiographic imaging with the machine and deep learning tools to characterize the mechanical damage in flaxseeds intelligently. Images were acquired under four levels of mechanical damage, and two strategies were used to discriminate seeds' damage: pattern recognition and convolutional neural network (CNN). In the former case, 69 morphological, color, and texture features were extracted. Various classifiers, namely, linear discriminant analysis (LDA), K-nearest neighbors (KNN), support vector machines (SVM), and decision trees were used for the analysis. SVM provided the best performance with a classification accuracy of 87.4%. Furthermore, the analysis of variance (ANOVA) F-test feature selection algorithm was utilized, and the 17 most effective features were selected to be used with an SVM classifier to classify seeds with 88.4% accuracy. In the case of CNN-based classifiers, six state-of-the-art architectures were employed including EfficientNet-B0, VGG19, Resnet18, MobileNet-v2, Inception-v3, and Xception. Among them, EfficientNet-B0 provided superior performance with a classification accuracy of 91.0%. The developed models' high accuracy confirms the capabilities of radiographic imaging and artificial intelligence tools for rapid, reliable, and automated assessments of mechanical damage in flaxseeds.