Media Produced by ASCII STARTUP
Media Produced by ASCII STARTUP
Tokyo, Shibuya Ward
Social infrastructure—such as tunnels, bridges, and concrete structures—is aging rapidly, and there is a growing shortage of skilled technicians to conduct inspections. In-Spector is developing an industrial diagnostic platform that combines hyperspectral imaging (HSI) with AI to visualize subtle signs of deterioration and damage—which are imperceptible to the human eye—in a non-destructive and non-contact manner. By analyzing multiscale spectral data obtained from satellites, drones, and ground-based sensors using machine learning, and by transferring the experiential knowledge of skilled inspectors to AI, the company aims to shift infrastructure maintenance toward preventive maintenance and reduce the need for human labor. As a DeepTech startup established in June 2025, it is promoting proof-of-concept (PoC) projects and joint research with companies and research institutions to facilitate real-world implementation.
[① Development of an HSI Foundation Model Architecture] Currently, training must be performed from scratch for each domain, requiring several months and several million yen to build a model for a single domain. We are developing a foundation model that learns general-purpose representations from large-scale unlabeled spectral data. This enables application to new domains using only a small amount of labeled data.
【② Multiscale Data Fusion Algorithm】 Technology to bridge the accuracy gap between high-resolution ground-based data (10–30 cm) and satellite data (30 m) has not yet been established. We will realize a scalable system that enables wide-area industrial diagnostics using only satellite data by linking ground and satellite pixels via GPS coordinates.
[③ Implementation of an Edge AI Inference System] Since current systems rely on cloud processing, there are barriers to on-site deployment in environments with unstable communication, such as inside tunnels. By combining model quantization, pruning, and knowledge distillation, we enable real-time on-site inference—from image capture to classification—in under several tens of seconds.
| Representative |
Jaewon Lee Currently pursuing a Ph.D. at the Graduate School of Interdisciplinary Information Studies, The University of Tokyo |
|---|---|
| Main Achievements | Participated in the 2026 ICT Startup League Won the 2024 S-Booster Award Selected for the 2024 Mitou Advanced Program |
| Company Name | In-Spector Co., Ltd. |
|---|---|
| Representative Name | Jaewon Lee |
| Address | 609 Miyamasuzaka Building, 2-19-15 Shibuya, Shibuya-ku, Tokyo |
| Official Site and Social Media, etc. |
Official Website |
| Founding Date | June 3, 2025 |