SEM-FCNET: SEMANTIC FEATURE ENHANCEMENT AND FULLY CONVOLUTIONAL NETWORK MODEL FOR REMOTE SENSING OBJECT DETECTION

It is challenging to detect objects in remote sensing images due to there being a large number of objects with few available features and a lot of background noise. Most existing methods ignore a large amount of background noise. In this paper, we propose an end-to-end based network model, Semantic feature Enhancement Model with a Fully Convolutional head prediction Network, referred to as SEMFCNet, to reduce the effect of background noise in remote sensing images detection. First of all, SEM-FCNet consists of a new semantic feature enhancement module to enhance the semantic features of small objects and reduce noise interference by fusing attention mechanism. Then, SEM-FCNet utilizes a fully convolutional head prediction network to detect multiple objects by extracting their location information. The experiments on two famous remote sensing image datasets, NWPU VHR-10 and HRSC, demonstrate the performance of the proposed SEM-FCNet model over the existing remote sensing image detection methods.