Automating Vessel Classification
With over 130,000 ships calling at Singapore every year, leveraging technology is key to monitoring our waters and guarding against maritime threats.
One of them, according to our technologists, is artificial intelligence. With the aim of augmenting sea situational awareness provided by traditional sensors such as radars, the multidisciplinary team collaborated with the RSN to develop a new command, control and communications (C3) system that would help the Maritime Security Task Force (MSTF) keep Singapore’s waters safe.
The C3 system uses video analytics to detect, track and classify vessels visually, reducing the cognitive workload of operators monitoring our waters. Development Programme Manager (C3 Development) Venessa Ng explained: “Using deep learning techniques, we were able to train video analytics models to infer a vessel’s class. This provides an added dimension to the C3 tactical picture, where vessel classification, typically carried out visually by the operator, is performed by the C3 system.”
An experienced operator usually takes around 20 to 120 seconds to identify and classify a single vessel. In contrast, the C3 system is able to identify and classify up to 20 vessels per second, thus relieving the operators of a significant cognitive burden and freeing them up to perform other important duties.
“The C3 system would be capable of maintaining oversight of vessels in Singapore’s waters on their behalf, and only alert them to anomalous observations that require special attention,” said Senior Engineer (C3 Development) Koh Jing Xuan.
The system’s success hinges on its video analytics performance – in particular, the accuracy of the machine learning models used for detecting and classifying vessels. Maximising this accuracy involved collecting large quantities of operational images to train the model, which then had to be labelled accurately one by one with help from RSN domain experts.
As engineers, the team decided to streamline this tedious process by developing their own data annotation software and scripts to automate the cleaning and processing of images and their labels.
Even before developing the C3 system, the team conducted rounds of interviews with the operators, and even shadowed them in their daily operations. With new insights into how the operators work and make use of information presented to them, the team was able to better identify tedious tasks in their workflow where automation would bring about the most value.
“At DSTA, we’re always exploring new applications of the latest tech, so once we heard about the requirements, we just gravitated towards using AI. We look forward to continue working with the RSN and the MSTF to enhance the system further,” said Head Capability Development (Navy C3 Development) Kang Shian Chin.
