Ship Target Detection Algorithm Based on Improved YOLOv3 for Maritime Image

Abstract

Accurate identification of ships is the key technology of intelligent transportation in water. At the same time, it also provides a judgment basis for water traffic safety control. This paper proposed a detection method of ships in water based on improved You Only Look Once version 3 (YOLOv3), which is called Feature Attention, Feature Enhancement YOLOv3 (AE-YOLOv3). The feature attention module was constructed by introducing the attention mechanism, which was embedded in Darknet-53 for feature recalibration, which improved the feature extraction ability of the model in the complex navigable background. For the problem of insufficient semantic information of low-level features in the feature fusion process, a feature enhancement module was constructed and applied to the feature fusion part to enhance the receptive field size of the corresponding feature layer and the correlation degree of feature extraction network. Experiments were carried out on the public SeaShips dataset. Experiments show that the detection accuracy is as high as 98.72%, which is better than that of other mainstream ship identification models, fully verifying the superiority of this method in the detection of waterborne traffic ships.

Research on ship recognition technology based on infrared image

“…r i P Class Object [68] to determine the probability that a target belongs to a class if there are targets in the grid. The final convergence result of each grid is the confidence of the target box, which can be obtained by formula (3) :…”

A Guide to Image and Video based Small Object Detection using Deep Learning : Case Study of Maritime Surveillance

“…Compared to the standard convolution, dilated convolution is more efficient since it enlarges the receptive field without increasing the number of parameters. Chen et al [181] proposed to enhance the feature representation of YOLOv3 by using multiple dilated convolutions to capture multi-scale context information for ship detection. Tian et al [166] embeded multiple Atrous Spatial Pyramid Pooling (ASPP) modules in FPN to improve the detection performance for ships at different scales.…”

A YOLOv3-Based Industrial Instrument Classification and Reading Recognition Method

“…After the images are annotated, the annotated file is divided into a training set and a test set, and the test set is fed into the YOLOv3 network [19][20][21][22] for training. e number of training rounds (epochs) is set to 200, the batch size of each iteration is 2, and the first 249 layers are unblocked so that they can be trained together.…”