DOI: 10.31015/jaefs.2026.2.22 ISSN: 2602-246X

Performance evaluation of deep learning models against IC-RT-PCR for accurate and cost-effective PVY detection

Mohammad Ehsan Alokozay, Zeynep Ünal, Çiğdem Serçe
Show PDF Cite
This study aims to develop DL (Deep Learning) models to compare their performance with the Immunocapture reverse transcription polymerase chain reaction (IC-RT-PCR) test for detecting Potato Virus Y NWI strain (PVY-NWI) in terms of accuracy, cost, and speed. Forty shoot samples were collected for both IC-RT-PCR and DL imaging. The images were captured directly from the field. In addition to comparing the laboratory methods with the DL method, six DL models were also compared. Six DL models were evaluated using both original and augmented image data. On the augmented data, the best model, with 99.65% accuracy, was ResNet101, indicating excellent performance and the potential to achieve 100% accuracy in future studies. The IC-RT-PCR method took approximately 17 h, 9 min, and 30 sec, while the DL models took a maximum of 12 sec and a minimum of 1.5 sec with augmented image data. This means the next time it will take only 1.5 seconds to detect PVY-NWI, whereas for the IC-RT-PCR method, the same time will be required, demonstrating the DL models' significant speed advantage. However, building the DL model requires a one-time investment of expenses and time. Using the model in the field for subsequent detections is highly cost-effective because it only requires a simple mobile phone camera for PVY-NWI detection. On the other hand, future runs of the IC-RT-PCR technique require the same reagents, labor, and time as the initial run.

More from our Archive