{"id":2060,"date":"2026-03-09T11:01:14","date_gmt":"2026-03-09T02:01:14","guid":{"rendered":"https:\/\/www.miyamoto-lab.com\/?page_id=2060"},"modified":"2026-03-17T04:19:32","modified_gmt":"2026-03-16T19:19:32","slug":"rock_en","status":"publish","type":"page","link":"https:\/\/www.miyamoto-lab.com\/ja\/rock_en\/","title":{"rendered":"AI-based Automatic Rock Particle Identification Algorithm"},"content":{"rendered":"\n

In fields such as crushed stone and aggregate management, embankment and excavated soil treatment, mining and quarry operations, and disaster mitigation, it is important to quantitatively characterize the properties of granular materials.<\/p>\n\n\n\n

The Miyamoto Laboratory has developed an original AI based algorithm that automatically identifies countless rock particles within seconds. The system generates detailed reports instantly, including grain size distributions and particle shape statistics.<\/p>\n\n\n\n

We are currently advancing the development of this system and provide consultation tailored to the needs of individual sites, from implementation to operational support.<\/p>\n\n\n\n

If you are interested, please feel free to contact us at: info@seed.um.u-tokyo.ac.jp<\/p>\n\n\n\n

<\/p>\n\n\n\n

Instant Extraction of Particle Features from a Single Image<\/strong><\/h5>\n\n\n\n

Our algorithm extracts the contour of each individual particle at the pixel level and outputs it as a polygon, rather than relying on simplified approximations such as rectangles or ellipses.<\/p>\n\n\n\n

This approach enables comprehensive characterization of particle properties. In addition to grain size, it provides detailed information on particle shape, including parameters such as aspect ratio, circularity, and convexity (Fig. 1).<\/p>\n\n\n\n

In terms of extracting information from rock particles, this represents the ultimate form of rock particle analysis.<\/p>\n\n\n\n

Fast, accurate, and comprehensive. Our technology extracts a complete set of particle properties from a single dataset.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 1. Automatic identification of particles using polygon representations and examples of particle parameters that can be derived from the results. The parameters shown here represent only a sample analysis; additional features can be implemented as needed.<\/p>\n\n\n\n

All You Need Is a Camera<\/strong><\/p>\n\n\n\n

No specialized sensors or large equipment are required. Our algorithm accurately identifies rock particles from virtually any captured image. The system works regardless of image resolution or color format, including both color and grayscale images.<\/p>\n\n\n\n

Images captured using smartphones, drones, fixed monitoring cameras, or other devices can be analyzed to automatically generate detailed reports, including grain size distributions and particle shape statistics (Fig. 2).<\/p>\n\n\n\n

The system supports both cloud and on-premise environments. By simply uploading images, users can obtain analysis results almost instantly. We also provide flexible support tailored to each site, from recommendations on imaging methods to the construction of analysis environments. We have already received inquiries from major construction companies.<\/p>\n\n\n\n

The system is also designed for operation in outdoor and harsh environments. For example, at the Hoei crater of Mt. Fuji, we integrated a compact camera and communication module into a weather-resistant enclosure and demonstrated stable operation even under limited power and communication conditions (Fig. 3).<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 2. AI-based automated granular analysis system developed by our team. The system can process a wide range of images, performs analysis instantly, and generates reports summarizing grain size distributions and particle shape distributions.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 3. Example of a field measurement system. The system was demonstrated to operate reliably even under limited power and communication conditions.<\/p>\n\n\n\n

A State-of-the-Art Algorithm Proven in Space Exploration<\/strong><\/p>\n\n\n\n

This algorithm was originally developed to address challenges identified during the JAXA asteroid exploration missions Hayabusa and Hayabusa2.<\/p>\n\n\n\n

When selecting safe and flat landing sites on asteroid surfaces covered with boulders, researchers previously had to analyze vast numbers of rocks manually, often working tirelessly to process the data.<\/p>\n\n\n\n

However, manual analysis faces fundamental limitations:
(1) enormous time and labor requirements,
(2) variability in results between analysts, and
(3) limited reproducibility even for the same analyst.<\/p>\n\n\n\n

Our algorithm resolves all of these issues.<\/p>\n\n\n\n

By analyzing approximately 10,000 images of the asteroids Ryugu and Bennu, the system successfully identified more than 3.5 million rock particles automatically. A scale of analysis that would be nearly impossible to achieve manually was completed in only about two weeks, including both preprocessing and postprocessing (Fig. 4).<\/p>\n\n\n\n

The results attracted significant attention and were widely reported through a University of Tokyo press release, appearing in more than 50 media outlets (Fig. 5).<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 4. Automated identification of rocks on an asteroid surface. A total of 3.5 million rocks were automatically identified using our AI based algorithm. By removing duplicates through map projection, we determined the locations and shapes of all rocks larger than 1 m, revealing the presence of approximately 200,000 such rocks, and derived key metrics including the size frequency distribution.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 5. Publication on the automated identification algorithm and media coverage. The work was featured by more than 50 media outlets and attracted broad attention.<\/p>\n\n\n\n

A State-of-the-Art Algorithm Proven in Space Exploration<\/strong><\/p>\n\n\n\n

This algorithm is currently being deployed in several space exploration projects.<\/p>\n\n\n\n

For example, in the initial analysis of samples returned from the asteroid Bennu by NASA\u2019s OSIRIS-REx mission in 2023, images of five sample trays provided through JAXA were analyzed within a few days. The algorithm successfully identified approximately 30,000 particles and revealed their grain size and shape distributions (Fig. 6). The speed and accuracy of this analysis have been highly evaluated by JAXA.<\/p>\n\n\n\n

The algorithm is also planned to be used in the upcoming Martian Moons eXploration (MMX) mission, scheduled for launch in 2026, and preparations for its application are currently underway. Interest from research institutions in the United States and Europe is also increasing, and the algorithm is attracting growing international attention.<\/p>\n\n\n\n

Importantly, the capabilities of the algorithm are equally powerful for applications on Earth.<\/p>\n\n\n\n

In 2022, we conducted a large-scale field demonstration at the Hoei crater of Mt. Fuji. Using drone imagery covering approximately 1.1 square kilometers, the algorithm automatically identified more than 8 million rock particles (Fig. 7).<\/p>\n\n\n\n

The analysis revealed not only grain size and particle shape distributions, but also the sources, trajectories, and displacement distances of falling rocks. These results provided new insights into the physical processes of rockfall and demonstrated that the algorithm can be applied reliably to terrestrial rock materials as well.<\/p>\n\n\n\n

The same algorithm operates with high precision both in space and on Earth. This universality represents one of the key strengths of our technology.<\/p>\n\n\n\n

<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 6. Example of the analysis of returned samples from asteroid Bennu. The analysis of five sample trays was completed within a few days.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Fig. 7. Example of large scale analysis at the Hoei crater of Mt. Fuji. Approximately 6,000 images were analyzed rapidly and comprehensively, enabling identification of all particles larger than several meters within the study area.<\/p>\n\n\n\n

Related links<\/strong><\/p>\n\n\n\n