Lightweighting aircraft type recognition technology for synthetic aperture radar imagery

Synthetic aperture radar (SAR) on a spaceborne platform enables the intelligent detection of aircraft in images and the identification of aircraft types, which is highly significant to the dynamic surveillance of sensitive areas, emergency rescue, situational analysis, and the acquisition of military intelligence. However, aircraft in SAR images present as discrete scattering points, which makes it challenging to effectively identify their types. Additionally, most networks are large and complex, which causes difficulty in applying them, owing to the limited computing power and storage space of spaceborne platforms. Therefore, this study proposes a lightweight aircraft-type recognition network for SAR imagery. The network utilizes FasterNet with a lightweight mechanism to construct the feature extraction module, which effectively reduces the model structure while ensuring the ability to extract features. In addition, a fast asymptotic feature pyramid network is employed that reduces the number of model parameters by gradually fusing feature information at each level. The effectiveness of each improvement module in the proposed method was evaluated using the SARAIRCRAFT- 1.0 aircraft type identification dataset, and the method was compared with other mainstream methods for verification. The results show that the developed lightweight network ensures high detection accuracy and aircraft category recognition capability while decreasing the number of model parameters by approximately 50.7%; hence, the number of parameters is significantly smaller than those of the comparison mainstream methods.