Focusing on the issues of low quality of image caption, insufficient utilization of image features and single-level structure of recurrent neural network in image caption generation, an image caption generation algorithm based on multi-attention and multi-scale feature fusion was proposed. The pre-trained target detection network was used to extract the features of the image from the convolutional neural network, which were input into the multi-attention structures at different layers. Each attention part with features of different levels was related to the multi-level recurrent neural networks sequentially, constructing a multi-level image caption generation network model. By introducing residual connections in the recurrent networks, the network complexity was reduced and the network degradation caused by deepening network was avoided. In MSCOCO datasets, the BLEU-1 and CIDEr scores of the proposed algorithm can achieve 0.804 and 1.167, which is obviously superior to top-down image caption generation algorithm based on single attention structure. Both artificial observation and comparison results velidate that the image caption generated by the proposed algorithm can show better details.

To extract richer texture features of face images to improve face recognition accuracy, a new face recognition algorithm based on the Shearlet_ULBP features which are extracted by the histogram of Uniform Local Binary Pattern (ULBP) from the Shearlet coefficients, called Shearlet_ULBP CRC (Shearlet_ULBP feature based Collaborative Representation Classification) was proposed. First, Shearlet transform was used to extract the multi-orientational facial information, and the average fusion method was exploited to fuse the original Shearlet features of the same scale. Second, the fused image was divided into several nonoverlapping blocks, and then face image was described by the histogram sequence extracted from all the blocks with the ULBP operator. Finally, the extracted features were fed into the collaborative representation based classifier. The proposed method can extract richer information about edge and texture features. Several experiments were conducted on the ORL, Extended Yale B and AR face databases, more than 99% recognition accuracy was achieved for images without occlusion, while the images are occluded, the recognition accuracy still reached more than 91%. The experimental results show that the proposed method is robust to the illumination, pose and expression variations, as well as occlusions.

In order to accurately detect the community structure of complex networks, a community detection algorithm based on signal adaptive transmission was proposed. First, the signal was adaptively passed on complex networks,thereby getting the vector affecting on the entire network of each node, then the topological structure of each node was translated into geometrical relationships of algebra vector space. Thus, according to the nature of the clustering, the community structure of the network was detected. In order to get the feasible spatial vectors, the optimum transfer number was determined, which reduced the searching space, and effectively strengthened the search capability of community detection.The proposed algorithm was tested on computer-generated network, Zachary network and American college football network. Compared with Girvan-Newman (GN) algorithm, spectral clustering algorithm,extremal optimization algorithm and signal transmission algorithm, the results show that the accuracy and precision of the proposed community division algorithm is feasible and effective.

In the process of remote sensing images fusion, the spectral distortion of fusion image is the main problem. To reduce distortion, an optimization image fusion method in combination with entropy component analysis transform was proposed. First, multi-band image was transformed to a small amount of bands by the entropy component analysis to reduce the spectral dimension. Projection transformation was finished from the perspective of entropy contribution so as to keep more information of source bands. Wavelet decomposition was done between the first entropy component and the high resolution image after histogram matching to get low frequency and high frequency subbands. For the fusion of low frequency subbands, Quantum-behaved Particle Swarm Optimization (QPSO) algorithm was applied to select the optimal weight coefficients. For the high frequency subbands, statistical feature and statistical model were used to perform fusion. The result of wavelet fusion was regarded as the first entropy principal component. The fusion image was obtained by wavelet and entropy component inverse transform. Entropy, cross entropy, standard deviation, grad, correlation coefficient and spectral distortion were selected as objective evaluation indexes. The experimental results show that the proposed method can enhance the spatial information and avoid spectral distortion.