Cattle Sperm Classification Using Transfer Learning Models

Abstract

This paper focused on classifying sperm and white blood cells (WBC) through image processing by utilizing different architectures of Transfer Learning Model (TLM). For image classification the researchers used microscopic images of sperm. A total of 602 image datasets were used for training and testing in deep learning with different convolutional network models. The models used are: InceptionResNetV2, Xception, DenseNet121, DenseNet169, MobileNetV1, InceptionV3, and DenseNet201. The classification of sperm and WBC is implemented successfully. The following is observed in the evaluation of these models: confusion matrix, loading time, weight size, and accuracy. From these evaluations: the highest model to consider for true positive is InceptionResnetV2. The accuracy of 98.3% is obtained by this model. However, the DenseNet121 also has comparable results with an accuracy of 95% considering its weight size of 93.49 MB as compared to InceptionResnetV2 of 641.93 MB.