Semantics-Consistent Representation Learning for Remote Sensing Image–Voice Retrieval

With the development of earth observation technology, massive amounts remote sensing (RS) images are acquired. To find useful information from these images, cross-modal RS image-voice retrieval provides a new insight. This paper aims to study task so as search effective data. Existing methods for rely primarily on pairwise relationship narrow heterogeneous semantic gap between and voices. However, apart included in datasets, intra-modality non-paired inter-modality relationships should also be taken into account simultaneously, since consistency among representations plays an important role task. Inspired by this, semantics-consistent representation learning (SCRL) method is proposed retrieval. The main novelty that takes pairwise, intra-modality, thereby improving learned SCRL consists two steps: 1) semantics encoding 2) learning. Firstly, image network adopted extract high-level features with transfer strategy, voice dilated convolution devised obtain features. Secondly, consistent space conducted modeling three kinds learn across modalities. Extensive experimental results challenging datasets show effectiveness method.