TY  - JOUR
T1  - AI-Based Model for Detection of Hazardous Chemicals Using Spectral Data Extracted Through Clustering
AU  - Ryoo, Seong-Min 
AU  - Kim, Yeon-Jin 
AU  - Cho, Sook-Kyung 
AU  - Baek, Sung-Ha 
AU  - Kim, Gyeong-Bae 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2024
DA  - 2024/1/1
DO  - 10.7840/kics.2024.49.8.1121
KW  - Detection of hazardous chemicals
KW  - Artificial intelligence
KW  - hyperspectral imaging
KW  - disaster and hazard
KW  - artificial intelligence training data
AB  - Hazardous chemical incidents can have a significant impact on the surroundings even with small quantities involved. Therefore, it is crucial to swiftly identify the chemical substance involved in an incident and respond appropriately when such incidents occur. Conventional studies on chemical substance identification have utilized chemical sensors and visual information. However, methods based on chemical sensors are challenging to apply when sensors are unavailable, and approaches relying on visual information face difficulties in detecting substances with identical or colorless characteristics. Therefore, this paper proposes a novel artificial intelligence-based hazardous chemical detection system using spectral spectrum data to address the challenge of distinguishing hazardous chemicals with the same color or those that are colorless. The paper introduces a technique for constructing artificial intelligence training data and a hazardous chemical detection model based on spectral spectrum data. The proposed artificial intelligence training data construction method improved the accuracy of data extraction by applying clustering to raw spectral spectrum data for extraction. Additionally, based on the extracted data and reflecting the characteristics of material spectral spectrum data, we proposed a hazardous chemical detection model implemented using the random forest algorithm. We validated the performance of the model through in-house experiments. The AI-based hazardous chemical detection system proposed in this paper is expected to minimize the damage from chemical incidents through rapid identification, even in cases where hazardous chemicals lack visual characteristics, enabling proactive response measures.