2.2. Model Setup

In this study, the datasets for training and testing were obtained following a cross-validation split stage. This stage will be discussed extensively in other sections. Pre-processing was performed on the training and testing datasets, consisting of several image augmentations. This augmentation aimed to increase the variation and number of images fed into the model, which would result in a performance boost [27,28]. Each of the augmentations applied to the images is discussed in the following section. The training dataset was used for the training stage to identify the labels between “sperm” and “impurity”. Once it completed its training, a final validation using the testing dataset was performed to predict the labels again. A visual representation of the experimental flow is shown in Figure 3. More detailed information on each experimental stage is provided in the corresponding sections below.

Experimental flow.