Identification of Maize Seed Defects using the Yolo-V5 Algorithmic Framework and Object Recognition Paradigms

Abstract

The maize quality, being one of the most extensively cultivated and planted crops across the globe, takes a central and significant role with respect to rendering farmers and the overall food sector prosperous and successful. Seed quality testing that is manually carried out is usually the norm, but it in turn renders it prone to multiple inefficiencies and chances of errors which can arise at the time of assessment. As land degradation continuing to impose serious risks upon agricultural activities, effective sorting technologies are increasingly essential and necessary to address these issues. Apart from this, maintaining proper and informed seed selection also heightens the production levels along with actively facilitating the practices of sustainable agriculture on a long-term basis. The system presented in this work employs several object recognition paradigms and the Yolov5 algorithmic framework to recognize faults in maize seeds. The work further verifies the performance of the developed model against major metrics of accuracy, precision, F1-score, and mAP. The proposed method employs computer vision algorithms for the recognition and classification of different seed defects, such as cracks, mildew, and discolored seeds. With a remarkable and impressive accuracy rate of up to 98%, the state-of-the-art system can successfully identify and classify abnormalities in seeds, provided it has already been trained on a large and comprehensive dataset that contains many annotated images of maize seeds. The state-of-the-art model called YOLOv5 achieved an extremely high F1-score of 97.7%, a result that reflects its incredible performance, as it was closely followed by another model, YOLOv8, which recorded impressive precision and mean Average Precision (mAP) results of 98.4% and 98.2%, respectively. These outstanding results go to verify and support the enormous potential and success of deep learning technologies in the accurate detection and recognition of defects in maize seeds.