OA-CapsNet: A One-Stage Anchor-Free Capsule Network for Geospatial Object Detection from Remote Sensing Imagery

Abstract

Object detection from remote sensing images serves as an important prerequisite to many applications. However, caused by scale and orientation variations, appearance and distribution diversities, occlusion and shadow contaminations, and complex environmental scenarios of the objects in remote sensing images, it brings great challenges to realize highly accurate recognition of geospatial objects. This paper proposes a novel one-stage anchor-free capsule network (OA-CapsNet) for detecting geospatial objects from remote sensing images. By employing a capsule feature pyramid network architecture as the backbone, a pyramid of high-quality, semantically strong feature representations are generated at multiple scales for object detection. Integrated with two types of capsule feature attention modules, the feature quality is further enhanced by emphasizing channel-wise informative features and class-specific spatial features. By designing a centreness-assisted one-stage anchor-free object detection strategy, the proposed OA-CapsNet performs effectively in recognizing arbitrarily-orientated and diverse-scale geospatial objects. Quantitative evaluations on two large remote sensing datasets show that a competitive overall accuracy with a precision, a recall, and an Fscore of 0.9625, 0.9228, and 0.9423, respectively, is achieved. Comparative studies also confirm the feasibility and superiority of the proposed OA-CapsNet in geospatial object detection tasks.