In the aviation field, accurate aircraft detection is of great significance for traffic monitoring, safety assurance, and military reconnaissance. Although traditional convolutional neural networks have achieved significant success in image recognition and object detection, they still face challenges in processing aerial images containing multi-scale and complex backgrounds. To address these issues, this study proposes a Vision Transformer based model that incorporates SKAttention (Selective Kernel Attention) and MSCAM (Multi Scale Channel Attention Module) technologies to improve the accuracy and efficiency of aircraft detection.
SKAttention technology effectively enhances the flexibility and accuracy of the model in processing aircraft images of different scales by adaptively selecting the most suitable convolution kernel size. MSCAM, on the other hand, optimizes the model's ability to process aircraft details and background information by enhancing channel attention at different scales. By combining these two methods into the Vision Transformer architecture, our model achieved accuracy and recall of 98.7% and 98.2%, respectively. These results validate the effectiveness of SKAttention and MSCAM in improving the performance of aviation aircraft detection based on Vision Transformer, providing new technological approaches and research directions for aviation image processing and object detection.
Aircraft detection based on improved Vision Transformer
International Conference on Mechatronic Engineering and Artificial Intelligence (MEAI 2023) ; 2023 ; Shenyang, China
Proc. SPIE ; 13071
2024-02-28
Conference paper
Electronic Resource
English
Aircraft inspection based on improved Swin Transformer
SPIE | 2024
|Vision-based aircraft guidance
NTRS | 1993
|