Aquaculture is the practice of reproducing, increasing and yielding aquatic organisms, such as aquatic animals and aquatic plants, in confined water bodies, like ponds, lakes, rivers, oceans, etc. Fishery represents one of the activities of aquaculture. The key function affecting fisheries management is the fishing activity. Operations like locating and counting fish are used to enhance this practice. There is a strong demand for underwater fish identification for multiple uses in sustainable fisheries. Real time monitoring helps to improve fishing activities. Deep Learning Techniques are used to train the computer with the available existing image data, faster GPUs, and algorithms employed to detect, locate and classify various objects within an underwater image or video with high accuracy. A well-liked object detection model, namely YOLO (You Only Look Once), is renowned for its quickness and precision. This paper presents a state-of-the-art version of the YOLO model for detecting and counting fish from underwater images or videos. The primary objective is to develop a system for automatic fish detection using an advanced convolutional neural network YOLOv8 and compare the results with the ones of the YOLOv7 model. It is proved that YOLOv8s performs better than YOLOv7, as YOLOv8s achieves a mAP@0.5 of 0.964, a precision of 0.929 and an IoU of 0.7.