An artificial intelligence (AI)-based surveillance system introduced to detect forest fires early in Gyeongnam is repeatedly misidentifying non-wildfire situations as fires. Last year, about 8 out of 10 suspected wildfire cases detected in Gyeongnam were false alarms caused by light, clouds, or fog being misrecognized.

According to the "ICT Platform Error Detection Status for Wildfire Prevention" compiled by the Korea Forest Service on the 23rd, the total number of platform detections in the Gyeongnam region last year was 106,061. Among them, cases that actually detected smoke or flames amounted to 22,140, while the remaining 83,921 were classified as errors.

The error detection rate was recorded at 79%, which is 9 percentage points higher than the national average (70%) last year. This is the highest figure among all regions. Following Gyeongnam were Gyeonggi (75%), Gangwon (74%), Jeonnam (57%), Daegu (51%), Ulsan (50%), and Gyeongbuk (49%).

The wildfire prevention ICT platform is a system where AI analyzes CCTV footage used for wildfire surveillance in real-time to spot smoke or flames. Unlike the traditional method where humans continuously monitored screens, it captures wildfire signs using AI deep learning and automatically alerts officials or the control room.

The Korea Forest Service launched a pilot project for the wildfire prevention ICT platform in 2020. In Gyeongnam, Hamyang-gun was first included in the project in 2020 (Note: Contextually referring to the initial rollout), and starting the following year, the project's scope expanded to a nationwide metropolitan unit encompassing all cities and counties in the province.

As of last year, Gyeongnam had a total of 248 wildfire surveillance CCTVs, including 216 panoramic types monitoring wide areas from mountain peaks and 32 close-up types checking specific locations such as trailhead entrances. Among these, 190 panoramic cameras are linked to the AI platform.

The problem is that there are many instances where AI fails to properly distinguish between wildfires and natural phenomena. In fact, light glare, foreign substances on lenses, clouds, fog, and even trees swaying in the wind are being misidentified as smoke or flames. So far, a total of 20.25 billion won—including 10.35 billion won in national funds and 9.9 billion won in local government funds—has been poured into this project through this year.

Kang Seung-kyu, a lawmaker of the People Power Party (Hongseong-Yesan, Chungnam) on the National Assembly's Agriculture, Food, Rural Affairs, Oceans and Fisheries Committee, stated, "There must not be situations where on-site response capacity drops due to frequent error detections." He added, "I urge the Korea Forest Service to conduct a full-scale inspection of the national platform's analysis performance and CCTV linkage status, and devise system improvement measures."

Rather than building additional platforms, the Korea Forest Service plans to focus on increasing the accuracy of existing systems. An official from the Wildfire Prevention Division under the Forest Disaster Control Officer of the Korea Forest Service said, "Platform operators are carrying out additional AI learning and program upgrades. The Korea Forest Service also plans to analyze the causes of errors with experts and companies to improve performance."

The official added, "There are still areas where accuracy falls short, such as detecting clouds and fog in addition to smoke and flames. To increase accuracy, camera performance and analysis technology must be improved together." They also noted, "While we continue AI learning and program upgrades, there are still shortcomings. We will look into the causes of frequent errors and discuss performance improvement measures with experts and companies."

[Copyright (c) Global Economic Times. All Rights Reserved.]