Deep learning-based approach using X-ray images for classifying Crambe abyssinica seed quality

The application of imaging technologies combined with state-of-the-art artificial intelligence techniques has provided important advances in the modern oilseed industry. Innovative tools have been designed to improve the characterization of different classes of seeds, and consequently, decision making has become more efficient. This study aimed to assess the potential of deep learning models based on convolutional neural networks (CNN) for monitoring the quality of crambe seeds using X-ray images. In the proposed approach, seeds with different physical and physiological attributes were used to create the models. The models achieved accuracies of 91, 95, and 82 % for discrimination of seeds based on the integrity of internal tissues, germination, and vigor, respectively. Therefore, our findings indicated that digital radiographic images are suitable to provide relevant information on the physical and physiological parameters of crambe seeds. Furthermore, the proposed methodology could be used to classify seeds quickly, non-destructively, and robustly.

Automated summary

The combination of imaging technologies and artificial intelligence has resulted in important advancements in the modern oilseed industry, leading to more efficient decision making. Deep learning models based on CNNs have shown potential in monitoring the quality of crambe seeds using X-ray images. The study demonstrated that digital radiographic images can provide relevant information on the physical and physiological parameters of crambe seeds, allowing for quick, non-destructive, and robust seed classification.