Small Object Detection (SOD) is a challenging task due to the small size of objects and the complexity of noisy backgrounds, which are common in fields like surveillance and autonomous driving. Traditional detection methods often suffer from information loss, poor feature representation, and localization errors, leading to reduced accuracy. This paper presents a novel approach that integrates knowledge graphs to enhance semantic consistency in small object detection. By quantifying relationships between objects and their environments, the proposed framework refines object predictions using contextual cues. Moreover, we develop a new core model based on YOLOv9, incorporating Space-to-Depth (SPD) layers and Convolutional Block Attention Modules (CBAM) to preserve fine-grained details while focusing on critical regions, significantly boosting detection accuracy. Extensive experiments on benchmark datasets demonstrate that the knowledge-assisted method delivers substantial improvements in average precision and robustness for detecting small objects in practical environments.