Bird and drone detection and tracking is a significant problem in air security. These targets are usually displayed far away with small sizes, making it difficult to obtain sufficient visual identification information.The latest detectors use multi-scale feature pyramids to deal with each size of the target, but still cannot handle them well when dealing with extremely small objects. Deep features of the lower layers emphasize visual information, while higher-resolution features of the upper layers obtain information which is useful for location search. Designing a well-structured deep neural network to complement these two features is an important issue in object detection.This paper proposes a novel model based on YOLO26-P2 [1-3] for detecting micro aerial targets. Using the detection ability of the higher-resolution feature of P2/4 layer as its base, this paper proposes a P3-to-P2 static fusion model and evaluates whether it improves the P2 feature. Next, this paper proposes a modified P3-to-P2 semantic feedback model and evaluates its performance by comparing it with the static fusion model. Unlike the static fusion model, the proposed method re-inserts the modified P3 feature into the P2 layer. This method is represented in the following form:P2 → Modified P3 → Modified P2This paper sets four models — the standard model [4, 5], the high-resolution P2 model, the P3-to-P2 static fusion model and the P3-to-P2 semantic feedback model, and compares each model to evaluate the performance of the proposed models.Finally, the ByteTrack [21] tracking algorithm was combined with the final detector to implement MOT(multi-object tracking).The proposed method applied simple changes of the feature layers within YOLO26 without any external image processing.In this paper, we set up a testing plan considering both the computational load and real-time performance to evaluate the performance of the proposed method.