Precise monitoring of Matsumurasca onukii is essential for data-driven integrated pest management (IPM), yet existing automated monitoring systems are often limited by unstable outdoor illumination, morphological degradation of trapped insects, and target occlusions, resulting in reduced detection accuracy. To address these challenges, an automated Internet of Things (IoT)-based tea pest monitoring system was developed through the coordinated design of an intelligent monitoring device and an enhanced detection model. The monitoring device employed an active mechanical decoupling mechanism to acquire images in a standardized optical chamber, while the detection model was designed to address detection difficulties associated with post-capture morphological degradation. In addition, a spatiotemporal correction procedure was introduced to improve counting reliability. Field trapping comparisons showed higher observed capture density for the IMD than for the standard and resized controls, with mean paired differences of 0.225 (95% CI, 0.109–0.380) and 0.275 (95% CI, 0.169–0.412) individuals/cm⁻², respectively, while RVBE-YOLO achieved an F1 score of 92.90%. Year-round deployment of three IMDs at one research site in 2025 showed close correspondence between system-generated and expert-derived daily counts (R² > 0.99; RMSE = 0.17–0.63) and captured clear seasonal variation in daily M. onukii trap captures. These results indicate that the proposed system can provide reliable and continuous records of M. onukii captures under the evaluated field conditions, supporting long-term automated pest monitoring in tea plantations.