Research on Multi-scale Object Detection Model Based on Transformer

In this paper, a Transformer-based multi-scale object detection model is proposed to solve the limitations of traditional object detection methods in multi-scale object detection. With its powerful self-attention mechanism, the Transformer model can effectively capture target features at different scales in the image. To verify the effectiveness of the model, this paper uses the COCO dataset to conduct extensive experimental studies. By introducing the multi-scale feature fusion strategy, this study achieves accurate target detection under multi-scale conditions. During the experiment, the model was gradually optimized, including adjusting the parameters of the multi-head attention mechanism and improving the Feature Pyramid Network (FPN) to better adapt to the needs of multi-scale detection. Finally, the experimental results show that the average accuracy (AP) of the proposed model on the COCO validation set reaches 41.2%, of which the detection accuracy (APs) of small-scale targets is 29.7%, and the AP values of 50.4% and 62.1% are obtained in the detection of medium-scale and large-scale objects, respectively. Compared with the traditional Faster R-CNN model, the detection performance of this model has been improved at all scales. In addition, the efficiency of the model was evaluated, and the results showed that the model could converge and achieve stable detection results in a short training time. These results demonstrate that the designed Transformer-based multi-scale object detection model achieves an effective balance between accuracy and efficiency, highlighting its significant potential for practical applications.