Submarine Pipeline Identification in Side Scan Sonar Image

Traditional image identification method is difficult to identify submarine pipelines covered by sand or sediments. To solve this problem, we propose a new method by combining BP neural network and genetic algorithm. In the proposed algorithm, BP neural network can be trained and then used to classify pipelines initially. To improve the accuracy of training model and prevent overfitting, we adopt multiple methods to expand data set, such as affine transformation, nonlinear transformation and so on. After initial pipelines fitting with BP neural network, we use genetic algorithm to fit the pipeline center with line. This line with the smallest residual sum of squares is where the pipeline lies. The experiments are conducted and comparing with the method based on Hough transform, the method proposed in this paper does better in identification accuracy and anti-interference.