Investigating the Effect of Spatial Fusion of Different Color Spaces on Deep Learning-Based Sperm Image Classification Performance

Image classification is a key application area of deep learning. The use of different color spaces in deep learning classification is also an important method employed to improve success rates. In this study, the HuSHeM dataset, composed of sperm samples defined in the RGB colors pace, was used to measure the impact of different color spaces on classification performance. The original RGB images were processed to obtain the LAB, HSV, and YCbCr color space equivalents of the same dataset. Subsequently, images defined in these different color spaces were spatially combined, and the effect of the implemented approach on deep learning-based classification was examined. In the obtained deep learning classification results, it was observed that the spatially combined images provided better classification success compared to the images in the original RGB color space and those in the other individual color spaces created. As a result of training the RGB, HSV, LAB, YCbCr, and combined images under the same conditions, classification accuracies of 73.16%, 72.92%, 73.64%, 69.2%, 74.32% were respectively achieved.