Object detection of inland waterway ships based on improved SSD model
Yang Yang, Pengyu Chen, Kaifa Ding, Zhuang Chen, Kaixuan Hu
Abstract
To realise the rational utilisation of inland waterway resources, the intelligent identification method based on Convolutional Neural Network (CNN) is used to track and monitor the ships. Introducing the Repulsion Loss function and Soft-NMS algorithm to improve model, improve the detection precision of the partially occluded ships. The Feature Pyramid Networks (FPN) is used to realise the fusion of semantic information and spatial information of feature map to solve the problem of difficult detection of small object ships. Three up-sampling methods are used to extend and smooth the feature map. Through the above multiple algorithm improvements, the partial occlusion ships and small object ships in inland waterways are effectively detected.