Convolutional neural network foresight sonar image recognition technology based on data increment

The invention discloses a forward-looking sonar image recognition technology of a convolutional neural network based on data increment, and the technology comprises five different underwater moving targets which are respectively a single-column target, a double-column target, a three-column target, a four-column target and a T-shaped target. The convolutional neural network foresight sonar image recognition technology comprises the following steps: step 1, extracting a training set and a test set from a foresight sonar original target image; 2, the target image is cut in the middle, and gray scale rotation is carried out; step 3, data increment; 4, inputting a convolutional neural network for training; step 5, obtaining a trained convolutional neural network; step 6, identifying and classifying the test set; according to the method, tedious feature engineering is not needed, the labor cost can be greatly reduced, the generalization ability is higher, and the training speed is higher. 本发明公开了基于数据增量的卷积神经网络前视声呐图像