Identification of haploid and diploid maize seeds using hybrid transformer model

Increasingly, more effective breeding techniques for new variations are preferred due to population growth and climatic change, particularly the accurate identification of the target variety. Maize haploid breeding technology, which can shorten the reproductive period and improve germplasm, has become the key to new maize breeding. In this study, a method in which deep features and image patches are analyzed together was proposed using a dataset consisting of 3000 different haploid/diploid type maize seed images in total. To achieve this objective, we adopted convolutional neural networks (CNNs) to recognize haploid and diploid maize seeds automatically through a transfer learning approach. More specifically, DenseNet201, ResNet152, ResNetRS50, RegNetX002, EfficientNetV2B0, EfficientB0, EfficientB1, EfficientB2, EfficientB3, EfficientB4, EfficientB5, EfficientB6, and EfficientB7 were applied for this specific task. The proposed hybrid model is inspired by both transfer learning and vision transformers. The error, accuracy, f1-score, recall, precision, and AUC of hybrid proposed model were 0.1491, 0.9633, and 0.9712, respectively. The accuracy rate reached, and the proposed model requires less processing in terms of complexity, which reveals the need for further investigation of such hybrid models. On the other hand, with the results obtained, it has been revealed that the maize seeds can be separated as haploid and diploid with traditional methods can be done much faster and without the need for an expert decision.