Maritime Target Detection Method Based on Deep Learning
Huixuan Fu, Yuan Li, Yuchao Wang, L. Han · 2018
For the problem of low recognition rate of traditional digital image processing in actual maritime target detection, an improved maritime target detection method based on deep convolution neural network is proposed. The maritime targets (mainly including fixed-wing aircraft, surface ships and helicopters under the background of sea and sky) are marked to establish the target datasets. The detection framework of maritime target based on Faster RCNN is given and improved by using the Resnet to extract the feature of target, and using the batch normalization layer to optimize the Faster RCNN network, using online hard-examples-mining algorithm to improve the training process. The training and test experiments of the improved target detection method were completed. The experimental results show that improved method can identify the common maritime target effectively and has better detection accuracy and robustness compared with Faster RCNN.