Research On Ship Target Detection In Remote Sensing Image Based On Deep Learning

Ship detection plays an important role in military security and people’s livelihood.In the research of ship detection,the target detection based on deep learning has made significant breakthrough.But the detection of small and dense targets at sea is still the focus and difficult area for the current research.Whether at sea or at port terminals,the rapid identification and accurate positioning of ships can help improve the safety of maritime navigation,which is of great practical significance for monitoring maritime traffic and safeguarding national maritime rights and interests..The main research contents of this paper is as follows:(1)In order to highlight the feature information of the ships in remote sensing image,this paper performs image preprocessing from three aspects: geometric transformation,image denoising and image feature extraction.Geometric transformation uses image rotation and image cropping to increase the sample size of the datasets.Image denoising using a combination of median filtering and bilateral filtering reduces the interference information of water ripples,waves and pretzel noise around the ship.The features are extracted using Laplace operator to highlight the feature information of the ship.(2)In order to improve the detection effect of small and dense targets in ship remote sensing images,an improved YOLOv3 ship detection algorithm is proposed in this paper.For the combination of feature maps,the dense connection of dense networks is used instead of the summation of residual networks,which better preserves the feature information of the target.Using DIo U as the loss function increases the sensitivity of the model to the overlap between the prediction box and the truth box.The improved algorithm is compared with the representative algorithm on two datasets,which proves that the algorithm has a good detection effect on ship detection.(3)In order to further improve the accuracy and speed of ship detection,an improved YOLOv4 ship detection algorithm is proposed in this paper.Firstly,the algorithm lightens the PANet module in CSPDark Net53 network and introduces depthwise separable convolution to optimize the five-layer continuous convolution of this module.Then,the multi-scale feature fusion of shallow networks is satisfied by fusing multiple SPP network structures.Finally,the Swish activation function is used to replace the Mish activation function in CSPDark Net53 network to increase the generalization of the network.The improved algorithm has been widely tested on two datasets and compared with other target detection algorithms to prove that the algorithm achieved better results in terms of ship detection accuracy and speed.