Researchers develop a cost-effective AI method to count manatees, aiding conservation efforts.
The conservation of endangered species receives a technological boost as scientists at Florida Atlantic University (FAU) devise an artificial intelligence (AI) method that accurately counts manatee populations in real-time.
Counting challenges and AI solution
Counting manatees has long presented a challenge due to their herding behaviour, weather conditions, time of day and environmental factors that obscure their visibility. Water reflections can also hinder the counting process.
As such, traditional aerial surveys, though useful, are hampered by high costs, variable accuracy, and observer bias.
Addressing these challenges, FAU's researchers have implemented a deep learning-based crowd counting approach that utilises closed-circuit television (CCTV) imagery, providing a real-time population estimate.
This study, detailed in the Scientific Reports journal, leverages an Anisotropic Gaussian Kernel (AGK) to match the manatees' distinct shape, transforming generic surveillance images into accurate density maps.
Cost-effective labelling and broad implications
The team adopted a deep learning-based crowd counting approach to automatically count manatees in a specific area, based on images from a CCTV camera. Then, they used line-label based annotation with a single straight line to mark each manatee.
This FAU-developed method outperformed other baselines. It worked particularly well when the image had a high density of manatees in a complicated background.
"Our method considered distortions caused by the perspective between the water space and the image plane. Since the shape of the manatee is closer to an ellipse than a circle, we used AGK to best represent the manatee contour and estimate manatee density in the scene," explained senior author Xingquan (Hill) Zhu, Ph.D., an IEEE Fellow and a professor in FAU's Department of Electrical Engineering and Computer Science.
"This allows density map to be more accurate, in terms of mean absolute errors and root mean square error, than other alternatives in estimating manatees' numbers," he added.
Besides creating a method that surmounts previous counting obstacles, the researchers have also provided their dataset and source code to the public, fostering further research.
The neural network-based approach balances the cost of labelling against the efficiency of counting, offering a straightforward, high-throughput counting method with minimal labelling effort.
It underscores the potential of computational techniques to enhance our understanding of endangered species populations, leading to more informed conservation strategies.
