An advanced YOLO-based image processing framework for automated sperm cell detection

Detection of sperm cell is an extremely important procedure in medical diagnostics and fertility research. Given the need to identify sperm in an efficient manner, this paper offers a powerful image processing pipeline. This method starts with image grayscale conversion, which is used to simplify the image and reduce the complexity of color, which is followed by Gaussian blur and Wiener filter to remove noise and improve image quality. An optimal-threshold is obtained by thresholding to obtain the sperm cells out of the background with the binarization method of Otsu which ends by concluding. Alternatively, adaptive Havrda-Charvat entropy thresholding is also discussed to allow better accuracy under difficult circumstances, correcting to differences in the intensity of the image. The sperm separation is then carried out in the segmentation phase, in which the sperm cells are segregated out of non-relevant objects. The last one is an automatic system that can identify sperm cells with high accuracy, which allows quicker and more accurate analysis by deep convolutional neural network of YOLOv5s to be used in clinical and research applications. The offered technique is tested with the help of a collection of microscopic images, which proves its efficiency under various light conditions and morphological changes of sperm.