Aiming at the problems that the high-capacity steganography model based on encoding-decoding network has weak robustness and can not resist noise attack and channel compression， a high-capacity robust image steganography scheme based on encoding-decoding network was proposed. In the proposed scheme， encoder， decoder and discriminator based on Densely connected convolutional Network （DenseNet） were designed. The secret information and the carrier image were jointly encoded into a steganographic image by the encoder， the secret information was extracted by the decoder， and the discriminator was used to distinguish between carrier images and steganographic images. A noise layer was added between the encoder and the decoder； Dropout， JPEG compression， Gaussian blur， Gaussian noise and salt and pepper noise were used to simulate a real environment with various kinds of noise attacks. The steganographic image output by the encoder was processed by different kinds of noise and decoded by the decoder. Through training the model， the secret information could be extracted from the noise-processed steganographic image by the decoder， so that the noise attacks could be resisted. Experiment results show that the steganographic capacity of the proposed scheme reaches 0.45 - 0.95 bpp on 360×360 pixel images， and the relative embedding capacity is improved by 2.04 times compared to the suboptimal robust steganographic scheme； the decoding accuracy reaches 0.72 - 0.97， and compared with the steganography without noise layer， the average decoding accuracy is improved by 44 percentage points. The proposed scheme not only guarantees high embedding quantity and high coding image quality， but also has stronger anti-noise capability.

Train assisted driving depends on the real-time detection of train operating environment. There are abundant small targets in the images of train operating environment. Compared with large and medium targets， small targets with the proportion of less than 1% of original image have problems of high missed detection and poor detection accuracy due to low resolution. Therefore， a target detection algorithm based on improved YOLOv3 in train operating environment was proposed， namely YOLOv3-TOEI （YOLOv3-Train Operating Environment Image）. Firstly， k-means clustering algorithm was used to optimize the anchor to speed up the convergence of the network. Then， dilated convolution was embedded in DarkNet-53 to expand the receptive field， and Dense convolutional Network （DenseNet） was introduced to obtain richer low-level details of the image. Finally， the unidirectional feature fusion structure of original YOLOv3 was improved to bidirectional and adaptive feature fusion structure， which realized the effective combination of deep and shallow features and improved the detection effect of the network on multi-scale targets （especially small targets）. Experimental results show that compared with original YOLOv3 algorithm， YOLOv3-TOEI algorithm has the mean Average Precision （mAP）@0.5 reached 84.5%， which increased by 12.2%， and the Frames Per Second （FPS） of 83， verifying that this algorithm has better detection ability of small targets in images of train operating environment.

Aiming at the universal No-Reference Image Quality Assessment （NR-IQA） algorithms， a new NR-IQA algorithm based on the saliency deep features of the pseudo reference image was proposed. Firstly， based on the distorted image， the corresponding pseudo reference image of the distorted image generated by ConSinGAN model was used as compensation information of the distorted image， thereby making up for the weakness of NR-IQA methods： lacking real reference information. Secondly， the saliency information of the pseudo reference image was extracted， and the pseudo saliency map and the distorted image were input into VGG16 netwok to extract deep features. Finally， the obtained deep features were merged and mapped into the regression network composed of fully connected layers to obtain a quality prediction consistent with human vision.Experiments were conducted on four large public image datasets TID2013， TID2008， CSIQ and LIVE to prove the effectiveness of the proposed algorithm. The results show that the Spearman Rank-Order Correlation Coefficient （SROCC） of the proposed algorithm on the TID2013 dataset is 5 percentage points higher than that of H-IQA （Hallucinated-IQA） algorithm and 14 percentage points higher than that of RankIQA （learning from Rankings for no-reference IQA） algorithm. The proposed algorithm also has stable performance for the single distortion types. Experimental results indicate that the proposed algorithm is superior to the existing mainstream Full-Reference Image Quality Assessment （FR-IQA） and NR-IQA algorithms， and is consistent with human subjective perception performance.

Water level prediction is an auxiliary decision support for flood warning work. For accurate water level prediction and providing scientific basis for natural disaster prevention， a prediction model combining Modified Gray Wolf Optimization （MGWO） algorithm and Temporal Convolutional Network （TCN） was proposed， namely MGWO-TCN. In view of the shortage of premature and stagnation in the original Gray Wolf Optimization （MGWO） algorithm， the idea of Differential Evolution （DE） algorithm was introduced to extend the diversity of the grey wolf population. The convergence factor during update and the mutation operator during mutation of the grey wolf population were improved to adjust the parameters in the adaptive manner， thereby improving the convergence speed and balancing the global and local search capabilities of the algorithm. The proposed MGWO algorithm was used to optimize the important parameters of TCN to improve the prediction performance of TCN. The proposed prediction model MGWO-TCN was used for river water level prediction， and the Root Mean Square Error （RMSE） of the model’s prediction results was 0.039. Experimental results show that compared with the comparison model， the proposed MGWO-TCN has better optimization ability and higher prediction accuracy.

Concerning the problems of the difficulties in obtaining， the limitation of labels， and the high cost of labeling of shale gas reservoir data， a Multi-standard Active query Multi-label Learning （MAML） algorithm was proposed. First of all， with the consideration of the informativeness and representativeness of the samples， the preliminary processing was performed on the samples. Secondly， the sample richness constraints including attribute differences and label richness were added， on this basis， the valuable samples were selected and the labels of these samples were queried. Finally， a multi-label learning algorithm was used to predict the labels of the remaining samples. Through experiments on eleven Yahoo datasets， the MAML algorithm was compared with popular multi-label learning algorithms and active learning algorithms， and the superiority of the MAML algorithm was proved. Then， the experiments were extended to four real shale gas well logging datasets. In these experiments， compared with the multi-label learning algorithms： Multi-Label Multi-Label K-Nearest Neighbor （ML-KNN）， BackPropagation for Multi-Label Learning （BP-MLL）， multi-label learning with GLObal and loCAL label correlation （GLOCAL） and active learning by QUerying Informative and Representative Examples （QUIRE）， the MAML algorithm improved the average prediction accuracy of comprehensive quality of shale gas reservoirs by 45 percentage points， 68 percentage points， 68 percentage points， and 51 percentage points， respectively. The practicability and superiority of the MAML algorithm in the prediction of shale gas reservoir sweet spots are fully proved by these experimental results.

In the current detection of the malware family, the local features or global features extracted through the grayscale image of the malware cannot fully describe the malware. Aiming at the problem and to improve the detection effect, a malware detection method based on perceptual hash algorithm and feature fusion was proposed. Firstly, the grayscale image samples of malware were detected through the perceptual hash algorithm, and samples of specific malware families and uncertain malware families were quickly divided. Experimental tests showed that about 67% malwares were able to be detected by the perceptual hash algorithm. Then, the local features of Local Binary Pattern (LBP) and global features of Gist were further extracted for the samples of uncertain families, and the features of merging the above two features were used to classify and detect the malware samples by the machine learning algorithm. Finally, experimental results of the detection of 25 types of malware families show that the detection accuracy is higher when using the fusion feature of LBP and Gist compared to that when using a single feature only, and the proposed method is more efficient in classification and detection than the detection algorithm using machine learning only with the detection speed increased by 93.5%.

In the production of printing industry， using You Only Look Once version 4 （YOLOv4） directly to detect printing defect targets has low accuracy and requires a large number of training samples. In order to solve the problems， a defect target detection method for printed matter based on Siamese-YOLOv4 was proposed. Firstly， a strategy of image segmentation and random parameter change was used to enhance the dataset. Then， the Siamese similarity detection network was added to the backbone network， and the Mish activation function was introduced into the similarity detection network to calculate the similarity of image blocks. After that， the regions with similarity below the threshold were regarded as the defect candidate regions. Finally， the candidate region images were trained to achieve the precise positioning and classification of defect targets. Experimental results show that， the detection precision of the proposed Siamese-YOLOv4 model is better than those of the mainstream target detection models. On the printing defect dataset， the Siamese-YOLOv4 network has the detection precision for satellite ink droplet defect of 98.6%， the detection precision for dirty spot of 97.8%， the detection precision for print lack of 93.9%； and the mean Average Precision （mAP） reaches 96.8%， which is 6.5 percentage points，6.4 percentage points， 14.9 percentage points and 10.6 percentage points higher respectively than the YOLOv4 algorithm， the Faster Regional Convolutional Neural Network （Faster R-CNN） algorithm， the Single Shot multibox Detector （SSD） algorithm and the EfficientDet algorithm. The proposed Siamese-YOLOv4 model has low false positive rate and miss rate in the defect detection of printed matter， and improves the detection precision by calculating similarity of the image blocks through the similarity detection network， proving that the proposed defect detection method can be applied to the printing quality inspection and therefore improve the defect detection level of printing enterprises.

The existing three-dimensional shape local features are mostly lack of spatial structure information and only contain a single property. In order to solve the problems, a hierarchical feature extraction framework integrating topological connection information of three-dimensional shape was proposed to obtain the three-dimensional shape ring feature with shift invariance. Firstly, based on the low-level feature extraction of a three-dimensional shape, the local region of feature points was modeled by the way of the isometric geodesic ring, which could extract the middle-level feature containing rich spatial geometric structure information. Then, the middle-level feature was further abstracted by using sparse coding to obtain more discriminative high-level feature with abundant information. The obtained high-level feature was compared with the existing Scale Invariant Heat Kernel Signature (SI-HKS) in two tasks of three-dimensional shape correspondence and shape retrieval, and its accuracy was increased by 24.5 percentage points and 7.2 percentage points respectively. The experimental results show that the proposed feature has higher resolution and recognition than the existing feature descriptors.

Using random projection acceleration technology to project the Singular Value Decomposition (SVD) of higher dimensional matrices onto a lower subspace can reduce the time consumption of SVD. The singular value random projection compression operator was defined to replace the singular value compression operator, then it was used to improve the Fixed Point Continuation (FPC) algorithm and got FPCrp algorithm. Lots of experiments were conducted on the original algorithm and the improved one. The results show that the random projection technology can reduce more than 50% time consumption of the FPC algorithm, while maintaining its robustness and precision. The modified matrix completion algorithm based on random projection technology is effective in solving large scale problems.

To overcome the weakness of over-concentration when the population of Particle Swarm Optimization (PSO) is initialized and the search precision of basic PSO is not high, an Improved PSO (IPSO) for constrained optimization problems was proposed. A technique of Good Point Set (GPS) was introduced to distribute the initialized particles evenly and the population with diversity would not fall into the local extremum. Co-evolutionary method was utilized to maintain communication between the two populations; thereby the search accuracy of PSO was increased. The simulation results indicate that, the proposed algorithm obtains the theoretical optimal solutions on the test of five benchmark functions used in the paper and the statistical variances of four of them are 0. The proposed algorithm improves the calculation accuracy and robustness and it can be widely used in the constrained optimization problems.

In this paper, a new monocular vision system was proposed to improve obstacle detection capability of indoor mobile robot. In this system, firstly, the Hue, Saturation, Intensity (HSI) color space conversion of images was performed. Secondly, a small target threshold selection method was proposed to segment the images, which enhanced the precision of the image segmentation. Thirdly, the target scene matching method and target projection matching method were used to calculate the change of the target pixel and projection so as to judge whether the target is obstacles or ground graphs. The experimental results show that the monocular vision system is effective and feasible, and this system can be applied to the navigation for small indoor mobile robots.

To improve the performance of a text-independent speaker recognition system, the authors proposed a speaker recognition system based on linear log-likelihood kernel function. The linear log-likelihood kernel compressed the input cepstrum feature sequence of a speaker model by a Gaussian mixture model. The log-likelihood between two utterances was simplified to the distance between the parameters of Gaussian mixture model. Polarization identity was applied to obtain the mapping from a cepstrum feature sequence to a high dimension vector. Support Vector Machine (SVM) was used to train speaker models. The experimental results on National Institute of Standard and Technology show that the proposed kernel has excellent performance.