Aircraft Detection Based on Improved DETR

With the wide application of deep learning in computer vision, Transformer-based target detection methods, especially DETR, have become a research hotspot. However, DETR still faces low detection accuracy and efficiency problems when dealing with small objects and complex backgrounds. In order to enhance the performance of DETR in aircraft detection, this paper proposes an aircraft detection method based on an improved DETR architecture, which mainly consists of two innovative improvements: first, an ECA attention mechanism is introduced after CNN feature extraction to enhance the model's adaptive learning ability for important features; second, Swin Transformer is used to replace the encoder in the original DETR -decoder structure in the original DETR, using its local windowing self-attention mechanism to improve the computational efficiency and accuracy of the model. The experimental results show that the method proposed in this paper improves the detection accuracy and speed compared with the traditional DETR in the aircraft detection task, especially in the complex background with better robustness and generalization ability. The research in this paper provides an effective improvement idea for the Transformer-based target detection method and provides new technical support for high-precision target detection tasks in practical applications.