Cross-Modal Hashing Retrieval Based on Deep Residual Network

In the era of big data rich inWe Media,the single mode retrieval system has been unable to meet people’s demand for information retrieval.This paper proposes a new solution to the problem of feature extraction and unified mapping of different modes:A Cross-Modal Hashing retrieval algorithm based on Deep Residual Network(CMHR-DRN).The model construction is divided into two stages:The first stage is the feature extraction of different modal data,including the use of Deep Residual Network(DRN)to extract the image features,using the method of combining TF-IDF with the full connection network to extract the text features,and the obtained image and text features used as the input of the second stage.In the second stage,the image and text features are mapped into Hash functions by supervised learning,and the image and text features are mapped to the common binary Hamming space.In the process of mapping,the distance measurement of the original distance measurement and the common feature space are kept unchanged as far as possible to improve the accuracy of Cross-Modal Retrieval.In training the model,adaptive moment estimation(Adam)is used to calculate the adaptive learning rate of each parameter,and the stochastic gradient descent(SGD)is calculated to obtain the minimum loss function.The whole training process is completed on Caffe deep learning framework.Experiments show that the proposed algorithm CMHR-DRN based on Deep Residual Network has better retrieval performance and stronger advantages than other Cross-Modal algorithms CMFH,CMDN and CMSSH.

A COMPARISON OF THE RETRIEVAL OF ATMOSPHERIC TEMPERATURE PROFILES USING OBSERVATIONS OF THE 60 GHZ AND 118.75 GHZ ABSORPTION LINES

The Microwave Temperature Sounder-Ⅱ(MWTS-Ⅱ) and Microwave Humidity and Temperature Sounder(MWHTS) onboard the Fengyun-3 C(FY-3 C) satellite can be used to detect atmospheric temperature profiles. The MWTS-II has 13 temperature sounding channels around the 60 GHz oxygen absorption band and the MWHTS has 8 temperature sounding channels around the 118.75 GHz oxygen absorption line. The data quality of the observed brightness temperatures can be evaluated using atmospheric temperature retrievals from the MWTS-Ⅱ and MWHTS observations. Here, the bias characteristics and corrections of the observed brightness temperatures are described. The information contents of observations are calculated, and the retrieved atmospheric temperature profiles are compared using a neural network(NN) retrieval algorithm and a one-dimensional variational inversion(1 D-var) retrieval algorithm. The retrieval results from the NN algorithm show that the accuracy of the MWTS-Ⅱ retrieval is higher than that of the MWHTS retrieval, which is consistent with the results of the radiometric information analysis. The retrieval results from the 1 D-var algorithm show that the accuracy of MWTS-Ⅱ retrieval is similar to that of the MWHTS retrieval at the levels from 850-1,000 h Pa, is lower than that of the MWHTS retrieval at the levels from 650-850 h Pa and 125-300 h Pa, and is higher than that of MWHTS at the other levels. A comparison of the retrieved atmospheric temperature using these satellite observations provides a reference value for assessing the accuracy of atmospheric temperature detection at the 60 GHz oxygen band and 118.75 GHz oxygen line. In addition, based on the comparison of the retrieval results, an optimized combination method is proposed using a branch and bound algorithm for the NN retrieval algorithm, which combines the observations from both the MWTS-Ⅱand MWHTS instruments to retrieve the atmospheric temperature profiles. The results show that the optimal combination can further improve the accuracy of MWTS-Ⅱ retrieval and enhance the detection accuracy of atmospheric temperatures near the surface.