Research On Deep Hashing Algorithm Based On Pairwise Labels

Hashing method is widely used as a nearest neighbor search technique for large-scale image retrieval.However,current hashing methods always extract features manually,which can not preserve the semantic similarity accurately.Obviously,manual extraction is hard to be applied for the large scale data retrieval directly.In recent years,with the rapid growth of convolution neural network and deep learning,the abstract features can be extracted from the massive data through complex network structures,which has achieved great success in the fields of computer vision and speech analysis.The traditional feature extraction technology is facing great challenges due to the uniqueness of the feature region indexing of molecular structured images.We apply the deep convolution neural network for molecular structural image feature extraction,which can generate a new type of molecular fingerprint to predict the relationship of GPCR(G Protein-Coupled Receptors)related ligand molecules and human common diseases.On the basis,we propose deep hashing method based on the pairwise labels,where the effective semantic-preserved hash codes can be yielded for large scale data retrieval by the supervised hashing method.Finally,the multiple experiments on the public biological data set demonstrate the advantage of our proposed method compared to the traditional hashing methods.