10:08 AM - 10:30 AM
[MIS15-24] Microfossil DX: Initiatives for automated high-throughput classification of radiolarian fossils
★Invited Papers
Keywords:microfossils, deep learning, paleoceanography
In the study of microfossils, species classification is one of the inevitable tasks, which is performed by a specialist with a large amount of knowledge and experience, spending a great deal of time and effort using microscopes. However, there is a limit to the amount of work that can be done by such specialists alone, and there is also concern about the recent decrease in human resources. On the other hand, high-resolution data analysis is becoming mainstream, requiring a large amount of analysis. It is becoming difficult to meet all requirements by specialists alone any longer.
Artificial intelligence (AI) is expected to overcome this challenge. In recent years, AI has come to be used in various aspects of our lives, promoting the so-called DX (Digital Transformation). In microfossil research, experiments have been conducted on the automatic classification of various fossil groups using deep learning, one of AI's learning methods. Many papers have shown that AI can classify microfossils at the species level, but for practical use, in addition to improving classification accuracy, the efficiency and high throughput of image data acquisition are an issue.
In 2017, GSJ (Geological Survey of Japan, AIST) began technological development toward the practical application of an automatic microfossil classification and picking system using AI, and this was put to practical use in 2018. Named the “miCRAD system”, this system is capable of automatically classifying, picking, and accumulating microfossils of a specific species by using a computer-controlled automatic microscope equipped with micromanipulators and an AI-based classification function, and two units are in operation at GSJ. In 2022, GSJ introduced a virtual slide scanner (hereinafter referred to as "slide scanner") to achieve high-throughput, instantaneous acquisition of an even larger volume of microfossil image data. Slide scanners are a technology that is becoming popular mainly in the medical field, and can acquire digital image data of slides (virtual slides) and share them with users around the world via networks. In this presentation, we would like to explain the results of operational tests of these systems at GSJ and their future development, using radiolarians as an example.
Artificial intelligence (AI) is expected to overcome this challenge. In recent years, AI has come to be used in various aspects of our lives, promoting the so-called DX (Digital Transformation). In microfossil research, experiments have been conducted on the automatic classification of various fossil groups using deep learning, one of AI's learning methods. Many papers have shown that AI can classify microfossils at the species level, but for practical use, in addition to improving classification accuracy, the efficiency and high throughput of image data acquisition are an issue.
In 2017, GSJ (Geological Survey of Japan, AIST) began technological development toward the practical application of an automatic microfossil classification and picking system using AI, and this was put to practical use in 2018. Named the “miCRAD system”, this system is capable of automatically classifying, picking, and accumulating microfossils of a specific species by using a computer-controlled automatic microscope equipped with micromanipulators and an AI-based classification function, and two units are in operation at GSJ. In 2022, GSJ introduced a virtual slide scanner (hereinafter referred to as "slide scanner") to achieve high-throughput, instantaneous acquisition of an even larger volume of microfossil image data. Slide scanners are a technology that is becoming popular mainly in the medical field, and can acquire digital image data of slides (virtual slides) and share them with users around the world via networks. In this presentation, we would like to explain the results of operational tests of these systems at GSJ and their future development, using radiolarians as an example.