YOLO, which has an extremely fast architecture with a detection pipeline consisting of a single neural network, is among the popular deep learning models of recent times. Since YOLO is a real-time object detector, its accuracy and detection speed are very important. When the model is evaluated from this perspective, activation functions help improve the performance of the model. In this study, performance improvements were made by using alternative activation functions instead of the default SiLU activation function of the YOLOv8 model. Appropriate activation functions were selected to increase the performance of the YOLOv8s model. ReLU, LeakyReLU, ELU, Sigmoid and HardTanh functions were used in the improved models. Experimental results were obtained using the BCCD and Urine datasets with the improved YOLOv8s models. The results were evaluated with recall, precision and mean Average Precision (mAP) metrics and the effect of the change of the activation function on the original YOLOv8s model was observed. When the experimental results and evaluation metrics were examined, better results were obtained compared to the original YOLOv8s model.