Employing deep learning for automatic river bridge detection from SAR images based on Adaptively effective feature fusion

Automatic river bridge detection is a typical and valuable application for SAR image analysis. However, the background information of complex, there are many specious targets with similar features, such as road, ponds ridges, which usually cause false alarms. And current methods fail to handle these interference efficiently. Therefore, this paper applies deep learning proposes new algorithm, named Single Short Detection-Adaptively Effective Feature Fusion (SSD-AEFF). It can effectively reduce noisy information, achieve fast high-precision bridges in complex imagery. SSD-AEFF based on SSD, AEFF module innovated enhance multi-scale feature maps together effective Squeeze-Excitation (eSE) further fuse features decrease information. Additionally, Non-Maximum Suppression (NMS) used screen out redundant candidate boxes produce final result. Moreover, Gradient Harmonizing Mechanism (GHM) loss function introduced solve problem sample imbalance training process. Experimental results TerraSAR data compared existing baseline models demonstrate superiority proposed algorithm.