Multiple deep learning by majority-vote to classify haploid and diploid maize seeds

Abstract

The maize plant is a crucial global staple, integral to food security. To ensure sustainable maize production, the development of high-yielding and resilient maize varieties is essential. This study proposes a majority voting-based decision support system for classifying haploid and diploid maize seeds using deep features from Convolutional Neural Networks (CNNs). Key variables include the accuracy, sensitivity, specificity, F-score, and Matthew's correlation coefficient (MCC) of the classification models. Experimental results showed impressive performance with accuracy, sensitivity, specificity, F-score, and MCC values of 90.96 %, 94.53 %, 86.40 %, 92.15 %, and 81.96 %, respectively. These results underscore the efficiency of the proposed method in accurately distinguishing between haploid and diploid seeds. The implementation of this decision support system in agricultural practices can significantly reduce the labour-intensive and time-consuming task of manual seed classification by experts. This system provides a cost-effective solution compared to existing expensive and complex methods, enhancing productivity, quality, and sustainability in maize breeding programmes. The ability to rapidly and accurately identify haploid seeds accelerates the breeding process, contributing to the development of new maize varieties with desirable traits such as higher yields and disease resistance. Future research should explore the integration of this decision support system with automation and robotics to further streamline the seed classification process. Additionally, investigating the applicability of this approach to other crops could broaden its impact. Further studies should also focus on enhancing the resolution of maize seed images and utilising more advanced hardware to improve processing performance. Finally, expanding the dataset with diverse maize varieties could refine the model's accuracy and generalisability.