KJCKorea
Journal Central

With the recent rapid proliferation of IoT devices, massive connectivity based on IoT network (LTE Cat.M1, NB-IoT and so on) is becoming increasingly frequent. In particular, when a large number of devices attempt to connect simultaneously, Radio Resource Control (RRC) connection resources within the base station become exhausted, leading to overload conditions that block new device access. To address this issue, this paper proposes an AI-based intelligent Inactivity Timer optimization algorithm. To overcome the rigidity of conventional fixed-threshold methods, the proposed algorithm mathematically defines the relationship between base station resource availability and Quality of Service (e.g., drop rate) and optimizes system weights in real-time using a Deep Q-Network (DQN). Performance evaluation results demonstrate that the proposed AI-based intelligent hybrid control model enhances capacity compared to existing fixed and rule-based methods in massive IoT environments.

Plant leaf diseases threaten crop yield and quality, while manual diagnosis requires considerable time and expert knowledge. This paper proposes a lightweight plant leaf disease classification framework combining Contrast Limited Adaptive Histogram Equalization (CLAHE), EfficientNet-B0, and the Convolutional Block Attention Module (CBAM). CLAHE enhances lesion contrast, EfficientNet-B0 extracts hierarchical features, and CBAM emphasizes disease-relevant features. The proposed framework was evaluated using 20,638 images from 15 PlantVillage classes through repeated experiments with three random seeds (42, 52, and 62) and compared with MobileNetV2, ShuffleNetV2, and SqueezeNet1.1. Experimental results showed that the proposed model achieved the highest classification performance, with a mean accuracy of 99.860 ± 0.019% and a macro-F1 score of 99.868 ± 0.036%. These findings suggest that the proposed framework has potential as a lightweight model for plant leaf disease classification in resource-constrained computing environments.

Wildfires are increasing in complexity under changing climate conditions and expanding forest –community interfaces, creating a need for spatially explicit, locally tailored prevention and response planning. This study classifies wildfire-vulnerable areas at the administrative-district level using public data and GIS-based spatial analysis in Andong-si, Yecheon-gun, Uljin-gun, and Yeongdeok-gun in northern Gyeongsangbuk-do. Forstudy aims to classify wildfire vulnerable areas by administrative unit using public data and GIS-based spatial analysis, focusing on Andong-si, Yecheon-gun, Uljin-gun, and Yeongdeok-gun in the northern region of Gyeongsangbuk-do. For this purpose, wildfire occurrence history, ignition causes, burned area, forest road and road accessibility, meteorological conditions, population distribution, firefighting water facilities, and fire station and 119 safety center location data were integrated. All spatial datasets were standardized using the GRS80 Central Belt coordinate system, and each indicator was transformed into a 0–100 score through Min-Max normalization. A Composite Risk Index was then calculated by applying weights to wildfire history, spatial accessibility, response infrastructure, and meteorological and social factors. Based on the index, wildfire vulnerable areas were classified into high-risk, caution, and monitoring areas. The results showed that Andong-si and Yecheon-gun had higher wildfire risk around forest boundaries adjacent to farmland and residential areas, mainly due to open agricultural burning. In contrast, Uljin-gun and Yeongdeok-gun showed relatively high wildfire spread risk due to coastal winds, mountainous terrain, forest-adjacent roads, and tourist activities. This study provides basic data for establishing region-specific wildfire prevention and response strategies that consider ignition probability, spread potential, and initial suppression capacity.