Ship Classification Based on Improved Convolutional Neural Network Architecture for Intelligent Transport Systems

In recent years, deep learning has been used in various applications including the classification of ship targets inland waterways for enhancing intelligent transport systems. Various researchers introduced different algorithms, but they still face problems low accuracy and misclassification other target objects. Hence, there is a need to do more research on solving above prevent collisions waterways. this paper, we introduce new convolutional neural network algorithm capable classifying five classes ships, cargo, military, carrier, cruise tanker The game dataset, which public dataset originating from Kaggle, all experiments. Initially, pretrained models (which are AlexNet, VGG, Inception V3 ResNet GoogleNet) were order select best model based its performance. Resnet-152 achieved with an 90.56%, AlexNet lower 63.42%. Furthermore, was improved by adding block contained two fully connected layers, followed ReLu characteristics our training dropout layer resolve problem diminishing gradient. For generalization, proposed method also tested MARVEL consists than 10,000 images 26 categories ships. compared existing algorithms obtained high performance others, 95.8%, precision 95.83%, recall 95.80%, specificity 95.07% F1 score 95.81%.