Dense Oil Tank Detection and Classification via YOLOX-TR Network in Large-Scale SAR Images

Oil storage tank detection and classification in synthetic aperture radar (SAR) images play a vital role monitoring energy distribution consumption. Due to the SAR side-looking imaging geometry multibouncing scattering mechanism, dense oil tasks have faced more challenges, such as overlapping, blurred contours, geometric distortion, especially for small-sized tanks. To address above issues, this paper proposes YOLOX-TR, an improved YOLOX based on Transformer encoder structural reparameterized VGG-like (RepVGG) blocks, achieve end-to-end densely arranged areas of large-scale images. Based YOLOX, encoder, self-attention-based architecture, is integrated enhance representation feature maps capture region interest tanks distributed scenarios. Furthermore, RepVGG blocks are employed reparameterize backbone with multibranch typologies strengthen distinguishable extraction multi-scale without increasing computation inference time. Eventually, comprehensive experiments Gaofen-3 1 m dataset (OTD) demonstrated effectiveness well performance superiority YOLOX-TR mAP mAP0.5 60.8% 94.8%, respectively.