Intelligent identification and counting model method for artificial fish reefs in sonar image based on deep learning algorithm

The invention provides an intelligent identification and counting model method for artificial fish reefs in sonar images based on a deep learning algorithm, and the method comprises the following steps: 1, data collection: employing a shipborne Ocuus 750d multi-beam imaging sonar to collect sonar images and video data of an artificial fish reef region and a surrounding marine environment; the collected images are stored in a computer, and an image and video database for training, verification and testing is built; 2, data preprocessing: dividing the data into a training set picture, a verification set picture and a test set picture according to a proportion of 8: 1: 1, and then performing random enhancement on the pictures by using data enhancement methods such as de-noising and image contrast enhancement in a Yolov8 framework; and step 3, feature extraction and analysis are carried out, and a deep learning algorithm of a Yolov8pose detection model improved by MobileNetV3 is utilized to carry out identificati