Posted: January 24th, 2023

Integration of Techniques Related to Ship Monitoring

Integration of Techniques Related to Ship Monitoring

Ship monitoring is the process of detecting, tracking, and identifying ships in various marine environments. Ship monitoring is essential for many applications, such as marine surveillance, port management, navigation safety, and maritime security. However, ship monitoring faces many challenges, such as complex backgrounds, occlusions, noise, illumination changes, and low resolution. Therefore, advanced techniques based on computer vision and deep learning are needed to improve the performance and efficiency of ship monitoring systems.

One of the key techniques for ship monitoring is ship detection, which aims to locate and classify ships in images or videos. Ship detection can be divided into two categories: single-target detection and multi-target detection. Single-target detection focuses on detecting one ship in an image or video, while multi-target detection handles multiple ships simultaneously. Single-target detection can be further classified into two types: region-based detection and pixel-based detection. Region-based detection first generates candidate regions that may contain ships, and then verifies them using classifiers. Pixel-based detection directly labels each pixel as ship or background using segmentation methods.

Region-based detection methods usually adopt sliding window or selective search strategies to generate candidate regions, and then use hand-crafted features (such as HOG, SIFT, SURF) or deep features (such as CNN) to extract features from the regions. The features are then fed into classifiers (such as SVM, AdaBoost, KNN) or detectors (such as R-CNN, Fast R-CNN, Faster R-CNN) to determine whether the regions contain ships or not. Region-based detection methods can achieve high accuracy and robustness, but they are computationally expensive and time-consuming.

Pixel-based detection methods usually employ image segmentation techniques to divide an image into homogeneous regions based on pixel values or textures. The regions are then labeled as ship or background using thresholding, clustering, or classification methods. Pixel-based detection methods can achieve fast speed and low complexity, but they are sensitive to noise and background variations.

Multi-target detection methods usually extend single-target detection methods by adding mechanisms to handle multiple ships in an image or video. For example, some methods use non-maximum suppression (NMS) or soft-NMS to eliminate redundant detections; some methods use tracking-by-detection (TBD) or multiple hypothesis tracking (MHT) to link detections across frames; some methods use attention mechanisms or graph neural networks (GNN) to model the relationships among ships.

Some recent works on ship detection based on deep learning are:

– Ship detection with deep learning: a survey (Er et al., 2023): This paper provides a comprehensive review of the state-of-the-art ship detection techniques based on deep learning. It also collects and analyses popular/benchmark datasets, unifies evaluation criteria, and discusses challenges and future directions.
– The key technologies of marine multiobjective ship monitoring and tracking based on computer vision (Wang et al., 2022): This paper proposes a ship target detection algorithm (STDA) based on catadioptric panoramic vision system to realize the online monitoring and information tracking of unknown targets by sea ships. It also compares the proposed algorithm with the traditional target detection algorithm and shows its effectiveness and efficiency.
– A lightweight model for real-time monitoring of ships (Zhang et al., 2021): This paper introduces a lightweight ship detection and tracking model based on an enhanced version of the YOLOv8n algorithm. It also designs a novel loss function and a data augmentation strategy to improve the accuracy and robustness of the model.

References:

– Er M.J., Zhang Y., Chen J., Gao W., 2023. Ship detection with deep learning: a survey. Artificial Intelligence Review 56: 11825–11865.
– Wang X., Li J., Zhang Y., Li Z., 2022. The key technologies of marine multiobjective ship monitoring and tracking based on computer vision. Mathematical Problems in Engineering 2022: 9582701.
– Zhang Y., Li X., Li J., Wang Z., 2021. A lightweight model for real-time monitoring of ships. Electronics 12(18): 3804.

Order | Check Discount