Abstract
This article provides a comprehensive understanding of the image-based ship detector using computer vision technologies with deep learning. Several pre-trained object detection models, such as MobileNet, VGGNet, Inception, and ResNet, have been investigated by illustrating the network architectures. A group of pre-trained models has been proposed and examined by recognizing ships on the sea and in the bay area. The model testing and comparison procedure have also been performed by evaluating the performance matrix and comparing predictive results per model. The optimal model is then chosen with the additional tests in terms of capabilities of the ship detection using the satellite image streaming in the real world. Such a proposed ship detector can contribute to the development of smart ship operations and may further carve out the possibility for the automated shipping system with smart port management.
TopIntroduction
In the era of big data and artificial intelligence, the international logistics system has been developing rapidly towards the next generation of intelligent objects by incorporating the smart sea machine. A smart ship refers to the application of sensors, the Internet of Things (IoT), and other hyper-connected objects to automatically detect and obtain signals and information steams from the ship, the marine environment, logistics, ports, etc. Such a system requires a set of cutting-edge technologies, such as computational science, automatic control theory, big data, machine learning, and computer vision. The intelligent operation of the ship is carried out in aspects of navigation, management, maintenance and cargo transportation; therefore, the logistics system will be safer, more environmentally friendly, cost-efficient and more reliable. A variety of intelligent architectures has been proposed throughout the front-to-end transportation and ship management, including navigation, hull, cabin, energy efficiency control, cargo management, and integration operation. Nowadays, research communities and industrial professionals emphasize the functionality of smart ship projects, which will benefit the construction of smart logistics.
From the technical perspective, a smart ship is designed to empower the sea machines in terms of making decisions without human interactions by incorporating artificial intelligence, which is the core technology of self-driving and vehicle collision warning systems. A smart ship is such a self-driving vehicle that will be competent in sensing its surrounding environment and moving safely in the water, without human operations. A set of state-of-the-art techniques, such as image recognition, object detection, and computer vision, plays a crucial role in the development of smart vehicle objectives, which have attracted attention from the data science community. Most recently, studies on self-driving vehicles have experienced a substantial enhancement due to improvements in deep learning. Deep neural networks have emerged as a powerful tool for image recognition and object detection by incorporating computer vision technologies, which provide the technical advances of smart ships in terms of image classifications.
With the development of deep learning, image classification and object detection techniques have been widely applied in the construction of the smart port and the Unmanned Surface Vehicle (USV) technology, whereas an effective and rapid detection approach is essential for the safe operation of the USV and the port management. With the improvement of the accuracy and real-time requirements of ship detection and classification in the practical application, it is necessary to propose a ship image/video detection and classification method based on an improved regressive deep network. However, such works are still challenging due to the limited data for the model training process, the complexity of model selection, and the difficulty of the ground-truth test. Motivated by the current demand for smart ship operations, this chapter is proposed to focus on the challenges behind the implementation of a smart ship. The objectives of this chapter are listed as follows:
- •
investigating how deep learning works for constructing the smart ship operation system in terms of the implementation of image recognition and object detection models using satellite images.
- •
examining how the ship detector is built using a variety of pre-trained object detection models, such as MobileNet, VGGNet, Inception, and ResNet, along with their performances on detecting ships through the model training and testing procedure.
- •
discussing the capability of the proposed ship detector given a real-world test, which provides a comprehensive understanding of implementations of the ship detector from the industrial application perspectives.
Key Terms in this Chapter
Image Recognition: One of the most classical issues in computer vision, image processing, and object detection, which deals with determine whether or not an image contains specific objects, patterns, or features.
Unmanned Surface Vehicle: A water-borne vessel that is able to operate on the surface of the water without any human operators.
Computer Vision: An automation technology that makes computers to gain high-level understanding from images and videos throughout acquiring, processing, analyzing, and recognizing digital data by transforming visual images into numerical or symbolic information.
Smart Port: An innovative port that applies automation and information technologies by incorporating big data, IoT, artificial intelligence, and blockchain.
Smart Logistics: The large-scale autonomous operation of inland vessels and seagoing machines using information collected by sensors.
Deep Learning: A broad family of machine learning models based on neural networks. Typical deep learning models are deep neural networks, convolutional neural networks, recurrent neural networks, deep belief networks, and deep reinforcement learning.
Convolutional Neural Network: A typical deep learning model that is commonly used to image classification, object detection, natural language procession, and predictive analysis. Such a network structure is a regularized version of fully connected networks, which belong to the class of artificial neural network.