Deep learning algorithm-based method for evaluating silting state of artificial fish reef in sonar image

The invention provides a deep learning algorithm-based method for evaluating a silting state of an artificial fish reef in a sonar image, 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; 2, data preprocessing, wherein the data are divided into training set pictures, verification set pictures and test set pictures according to the proportion of 8: 1: 1; step 3, feature extraction and analysis: using a deep learning algorithm of a MobileNetV3 improved Yolov8pose detection model to identify the artificial fish reef, extracting and screening key point features of an artificial fish reef cube plane, and determining a feature mode related to the artificial fish reef so as to facilitate subsequent attitude state identification training; and 4, model construction: extracting a water bottom plane and an artificial fish reef silting par