Over the years, the rising concerns over marine pollution has irked scientists triggering research to find solution. To deduce better marine environment solutions, scientists need to know more about the depths of the seas. Till date, marine life researchers are exploring seas with Underwater Autonomous Vehicle (UAV) which has latest cameras and sensors mounted over it.
According to a new study, marine scientists can get 83% accurate data with the help of robotics. However, with further implementation of artificial intelligence and training the algorithm with enough data, scientists can get better results.
Nils Piechaud, a Ph.D. student and lead author of study states, UAV can explore a vast area of the sea at a depth of 60 m. However, manual processing of the vast data is virtually impossible. This makes AI a promising tool to identify various species that live down there. But there are chances that they might get wrong results every five times.
Google Tensorflow Helped Scientist to Have Better Results
Autosub6000, a British UAV collected more than 150,000 images in a single 1200 m dive. Out of these images, more than 1200 were manually analyzed. These 1200 images had approximately 40,000 specimens of 110 species. To train their AI model, scientists used Google Tensorflow. This predefined open access library helped Convolution Neural Network (CNN) to identify specimens from the data derived from AUV.
Although, the accuracy of human annotation is approximately 50%-80%, the process is extremely slow and gets quite inconsistent across time. On the other hand, with the implementation of AI, can improve the accuracy can to 85%.
Though the study does not prove that AI will replace manual annotation, still proves itself to be a handy tool to analyze marine life. Study demonstrates the implementation of AI can assist scientists to explore the untouched realms of marine life.