Classification of haploid and diploid maize seeds by using image analysis and machine learning algorithms

Doubled haploids is now widely used in advanced maize breeding programs. This technique shortens the breeding period and increases the efficiency of breeding. One of the important processes in this breeding technique is the selection of haploid seeds. The most common method of selecting haploid seeds is the R1-nj color marker. The fact that this selection is performed manually reduces the selection success and causes time and labor loss. For this reason, it has become a need to develop automatic selection methods that will save time and labor and increase selection success. In this thesis, a method is proposed to classify haploid and diploid maize seeds using image analysis and machine learning algorithms according to the R1-nj color marker. Within the scope of the thesis study, two separate datasets were created. As the feature vectors; 8, 16 and 32 bin grayscale color histograms, color moments and texture features were used. The obtained feature vectors is classified by decision tree, k-nearest neighbors, support vector machine and artificial neural network. The classifier performance was tested by a 10-fold cross-validation method. Best performance has been obtained with 90,39% accuracy rate by using the color moments features with the k-nearest neighbor classification method.