Image Ranking Relevancy Based on Semantic Web Using Deep Learning Technique

Abstract

Computer vision and deep learning have significant leverage on the retrieval of image ranking. The impressive advancements of deep learning techniques for computer vision and other applications conducted an excellent performance for semantically image ranking. The great challenge in image ranking task concentrates on extracting the deepest features of the image. This paper investigates a highly scalable and computationally efficient of deep relevance image ranking system for large scale images. The superior deep network model called RetinaNet is utilized as a feature extractor to learn deep semantic feature embedding of the imaging data. Besides, The effective transfer learning scheme is proposed to transfer the RetinaNet learning to deep relevance image ranking system. The experimental results manifest that our deep learning procedure enhancement the retrieval results efficiently and accurately and focuses on inhibit the learning time of a deep, relevant ranking task. As compared with other state-of-the-art object detectors, the RetinaNet detector accomplished more than a 97% mean average precision (MAP). These superior results pretend the effective impact of our proposed procedure learning that drives the more efficient and relevant result of the deep ranking task.