建设一流本科教育是大学内涵式发展的需要和目标,振兴本科教育,全面提高人才培养能力,走内涵式发展道路,是我国高等教育发展的必由之路.一流本科教育中的人才培养应以学生创新实践能力的培养为核心.本研究分析了当前关于学生创新实践能力培养的研究现状,探讨了计算机科学与技术专业的人才培养模式,提出"多维一体"的教学实施机制,对其中采用的"以赛促学、以用促学"的实践路径和激励措施进行了阐述,以期为大学生创新实践能力培养提供参考.

In order to enhance the feature extraction ability of neural network and further improve the accuracy of facial expression recognition, this paper proposes a deep separable residual network model under joint loss DSResNet-Jloss.This network is a lightweight network model based on deep separable convolution and residual learning methods.The method of channel-by-channel convolution and point-by-point convolution is used to replace the conventional convolution operation, which solves the problems of traditional convolutional neural network with large parameter redundancy, long training time, slow convergence, and easy overfitting.And add residual unit to the network, use shortcut connection, through identity mapping, to solve the problem of gradient explosion or attenuation caused by too many layers of the network model.A joint loss function is proposed, which fully combines the advantages of cross-entropy loss, center loss and contrast loss to reduce the intra-class distance of expression features and increase the inter-class distance.Experiments show that the model has achieved good results on the two public data sets of FERPlus and RAF-DB, showing good generalization ability and robustness.

Aiming at the problems of inaccurate facial expression recognition and large amount of calculation under multi-perspective in real life, a facial expression recognition model MVResNet-FER is proposed, which is based on multi-perspective feature fusion under deep residual convolution.The residual block in ResNet is first improved and the conventional convolutional network is replaced with a depthwise separable network.Second, a CBAM module is added to enhance the extraction of effective features under multi-perspective and the supplementation of shallow feature information. Then use the RReLu activation function to replace the original ReLu to avoid deactivation of some nodes when the gradient is large.Finally, the global average pooling layer is used instead of the fully connected layer to achieve dimensionality reduction, and the generated feature vector is sent to Softmax for classification Experiments show that the proposed method produces excellent results on the CK+ and RaFD datasets, which can effectively improve the accuracy of facial expression recognition.

In order to solve the problems of insufficient facial expression feature extraction and large parameter amount in some convolutional neural networks, a facial expression recognition algorithm based on multi-feature fusion is proposed. This method first modifies the residual block in the ResNet network, reduces the amount of network parameters and uses pre-activation to reduce the error rate. After that, the features extracted by the improved ResNet network are fused with the features extracted by the VGG network after the cut layer, and the network model P-ResNet-VGG is obtained. The loss function uses the cross entropy loss function. This model has been extensively tested on the FER2013 and JAFFE datasets. The experimental results show that this model has improved accuracy on the expression data set than other models, and it has a significant effect on the FER2013 and JAFFE data sets.

In order to further improve the accuracy of facial expression recognition in small samples, a small sample expression recognition method based on deep learning and fusion of different models is proposed. In this method, a single CNN model is first compared, and the relatively appropriate convolutional neural network (CNN) is selected by preserving probability of different nodes in the dropout layer. Then, the scale-invariant feature transformation (SIFT) algorithm is used to extract features. The purpose of extracting features with SIFT is to improve the performance of small data. And then, in order to reduce the error, avoid over fitting, all the model to carry on the summary, all the model of the weighted Average CNN-SIFT-AVG (Convolutional Neural Network and Scale Invariant Feature Transformation business) model. Finally, only a few sample data are used to train the model. The model has been tested on FER2013, CK+ and JAFFE datasets. Experimental results show that this model can greatly improve the accuracy of small sample facial expression recognition, and has produced excellent results in FER2013, CK+ and JAFFE dataset, with a maximum improvement of about 6

An image defogging algorithm based on attention mechanism and Markov discriminator (PatchGAN) is proposed herein to differentiate features according to the different regional features of foggy images, which cannot be achieved using existing defogging algorithms. Combined with attention mechanism, the proposed algorithm can adaptively assign weights to the features of different regions while using the module with Inception mechanism to predict the globally relevant atmospheric light value more accurately and effectively. The predicted atmospheric light value, transmittance, and foggy image are input into the atmospheric scattering model to obtain the defogged image. Finally, the defogging image is input into PatchGAN to determine whether it is true or false. The experimental results show that the proposed algorithm achieves good defogging effect on indoor and outdoor foggy images and improves the brightness and saturation of defogging images.

In order to further improve the accuracy of facial expression recognition, a deep convolutional neural network algorithm fusing global and local features (GL-DCNN) is proposed. The algorithm consists of two improved convolutional neural network branches, global branch and local branch, which are used to extract global features and local features respectively. The features of the two branches are weighted and fused, and the fused features are used for classification. Firstly, global features are extracted. The global branch is based on transfer learning, and the improved VGG19 network model is used for feature extraction. Secondly, local features are extracted. In the local branch, central symmetric local binary pattern (CSLBP) algorithm is used for the first feature extraction, and the local texture information of the original image is obtained, which is input into shallow convolutional neural network for the second feature extraction, so that the local features related to facial expressions are automatically extracted. Thirdly, two cascaded fully connected layers are used to reduce the dimension of the features of the two branches, and different weights are assigned to them for weighted fusion. Finally, softmax classifier is used for classification. The experiment is validated on CK+ and JAFFE datasets, and the classification accuracy is over 95% and 93%, respectively. Compared with other five algorithms, this algorithm has a good overall performance, good recognition effect and good robustness, which can provide an effective basis for facial expression recognition.

In order to solve the problem that the existing defogging algorithms can’t differentiate according to the characteristics of different regions of the fogged image, an Image Defogging Algorithm Based on Inception Mechanism is proposed(I-defog algorithm). The attention mechanism is added to the algorithm to adaptively assign weights to the features of different regions; It is more accurate and effective to use the module with perception mechanism to predict the global value. The predicted value, transmittance and foggy image are input into the atmospheric scattering model to get the defogging image, and the defogging image is input into the Markov discriminant (PatchGAN) to judge whether it is true or not. The results show that the algorithm achieves good defogging effect on both indoor and outdoor images, and improves the brightness and saturation of defogging images.

Aiming at the problem of low recognition rate caused by the lack of effective identity information in small-scale face images with low resolution, this small face recognition algorithm based on super-resolution reconstruction is proposed. The algorithm first performs super-resolution reconstruction on the collected low-resolution face images, and uses the method of detail enhancement to restore high-frequency information such as facial contour information and texture details of the image, and then uses an improved densely connected network to do feature extraction and image recognition. Experimental results show that this method is aimed at small-scale face images, and is superior to other face recognition algorithms in image recognition rate, and can effectively solve the problem of low recognition rate of small faces in real environments.

In order to solve the problem of single-channel convolutional neural network feature loss in the process of facial expression recognition, a facial expression recognition algorithm based on multi-feature fusion network is proposed. The algorithm uses the dual-channel convolutional neural network model DCNN-FER (Dual-channel Convolutional Neural Network Model for Facial Expression Recognition). The pre-processed face image is input to channel one to obtain global features, and the face image that has been processed by Local Binary Patterns (LBP) is input to channel two to obtain local texture features. At the same time, it is used in part of the convolutional layer. The Convolutional Block Attention Module (CBAM) enhances the network’s focus on the useful information of the image and suppresses useless features. Finally, new features are formed by weighted fusion and sent to the softmax layer for classification. This algorithm not only considers the extraction of overall facial features, but also enriches local texture features. Compared with other methods on the FER2013 and CK + facial expression data sets, the method in this paper shows good robustness.

Aiming at the problem that the feature vector extracted by original Gabor transform has a high dimension and DBN ignores local information when completing face recognition, a DBN face recognition algorithm based on GCSLBP is proposed. Firstly, the algorithm improves the original Gabor transform, and optimizes it by introducing the LBP . Then the algorithm uses the histogram method to represent the final feature vector, which extracts the rich local features of the image and reduces the dimensionality of the feature vector. Finally, DBN is used to improve the robustness of classification and complete the classification and recognition of faces. The algorithm has been simulated experiments on ORL and CMU_PIE data sets. The experimental results show that the algorithm in this paper effectively improves the face recognition rate and are also robust under light and other transformations.

针对图像特征提取方法提取单一特征不能很好地表示图像的问题,提出了二维线性鉴别分析和协同表示的面部识别方法.该方法首先通过二维线性鉴别分析(Two-Dimensional Linear Discriminant Analysis,2DLDA)分别对训练样本的类间散布矩阵和类内散布矩阵提取特征,之后利用得到的特征重建图像,包括类间虚拟图像和类内虚拟图像.其次,将类间虚拟图像、类内虚拟图像和原始图像利用协同表示(Collaborative Representation,CR)算法进行得分.最后,采用加权得分融合算法将上述得分进行融合以获得最终得分,并利用最终得分进行图像识别.该方法不仅有效的抑制了光照和表情对面部识别的影响,同时根据获得的类间虚拟图像、类内虚拟图像与原始图像互补,有效的提高面部图像识别的性能.实验结果表明,该方法在不同的数据库下(ORL、AR、GT)具有较好的识别精度.

In order to solve the problem that the face image is not rich in features extracted under complex lighting environments，which leads to a low recognition rate，a local pattern feature extraction and recognition algorithm based on sparse representation is proposed. Firstly，the image is divided into several sub-images and the Dynamic Threshold Central-symmetric Local Binary Pattern ( DTCLBP) algorithm is used to extract features by thresholding the pixels of each sub-block and encoding the results of comparison with the central pixel values into the Central Symmetric Local Binary Pattern ( CSLBP) ; and then second-order features are extracted from the processed image by the former step using the Central Symmetric Local Derivative Pattern ( CSLDP) ; finally，the sparse representation classification algorithm is used to classify and identify the extracted features. The simulation experiments on Extended Yale B，CMU _PIE and AＲ face databases validate the effectiveness of the DTCLBP-CSLDP-SＲC.

面对疫情以来的线上教学实践,探讨黑龙江省高校在线教学的发展状况,从计算机类课程的基本特点出发,分析传统课堂教学的单向性、封闭性等教学方法的不足.结合线上教学的特点和优势,提出改变教学观念,遵循以学生为中心的自主学习教学理念.结合现代教学技术,介绍线上线下混合式教学模式的构建,以重建后疫情时代计算机教学的解决方案.

Recently, cross-view feature learning has been a hot topic in machine learning due to the wide applications of multiview data. Nevertheless, the distribution discrepancy between cross-views leads to the fact that instances of the different views from same class are farther than those within the same view but from different classes. To address this problem, in this paper, we develop a novel cross-view discriminative feature subspace learning method inspired by layered visual perception from human. Firstly, the proposed method utilizes a separable low-rank self-representation model to disentangle the class and view structure layers, respectively. Secondly, a local alignment is constructed with two designed graphs to guide the subspace decomposition in a pairwise way. Finally, the global discriminative constraint on distribution center in each view is designed for further alignment improvement. Extensive cross-view classification experiments on several public datasets prove that our proposed method is more effective than other existing feature learning methods.

In order to further improve the accuracy of facial expression recognition in small samples, a small sample expression recognition method based on deep learning and fusion of different models is proposed. In this method, a single CNN (convolutional neural network) model is compared, and the relatively appropriate CNN is selected by preserving probability [p] of different nodes in the dropout layer. Then, the scale-invariant feature transformation (SIFT) algorithm is used to extract features. The purpose of extracting features with SIFT is to improve the performance of small data. And then, in order to reduce the error, and avoid over fitting, all the models are carried on summary，and the model CNN-SIFT-AVG (convolutional neural network and scale-invariant feature trans-formation average) is obtained by simple average model fusion method. Finally, only a few sample data are used to train the model. The model is tested on FER2013, CK+ and JAFFE datasets. Experimental results show that this model can greatly improve the accuracy of small sample facial expression recognition, and produce excellent results in FER2013, CK+ and JAFFE datasets, with a maximum improvement of about 6% compared with other facial expression recognition methods.

针对人脸图像中不同部位所含的信息熵不同,对识别的影响程度不同等因素,提出了一种信息熵加权的HOG特征提取方法.该算法将待识别的人脸图像进行分块,对分块后的图像进行HOG特征提取,计算每块图像所含的信息熵作为权重系数加到各个分块中形成新的HOG特征,通过PCA算法对特征进行降维,得到信息熵加权的HOG特征.通过在ORL和YALE实验结果表明,该算法相较于其他传统识别方法具有更高的识别精度和准确度,并且对于人脸在光照、姿态表情等干扰因素下均具有良好的有效性和鲁棒性.

The purpose of multiview clustering is to improve clustering performance by using multiview information. At present, the research of multiview mainly focuses on the multiview similarity learning. However, the existing multiview similarity learning methods have two drawbacks. First of all, similarity learning is learned based on the fixed graph learned from each view separately, thus ignoring the potential relationship between the views. Secondly, the real data is often destroyed. When the data are contaminated with noise or outlier, the learned similarity graph is unreliable. Therefore, this paper proposes a Reliable Multiview Graph Learning method(RMGL). This method can deal with noisy data. By constructing a data recovery model to restore the true distribution structure of data from each view, the reliability of similarity learning could be guaranteed. Then, a reliable neighbor similarity is obtained with multiview adaptive graph learning. Experimental results on several datasets show that the proposed method is superior to the existing methods on clustering performance and has strong robustness to noisy data.

In order to improve the detection rate of the traditional single-shot multibox detection algorithm in small object detection, a feature-enhanced fusion SSD object detection algorithm based on the pyramid network is proposed. Firstly, the selected multiscale feature layer is merged with the scale-invariant convolutional layer through the feature pyramid network structure; at the same time, the multiscale feature map is separately converted into the channel number using the scale-invariant convolution kernel. Then, the obtained two sets of pyramid-shaped feature layers are further feature fused to generate a set of enhanced multiscale feature maps, and the scale-invariant convolution is performed again on these layers. Finally, the obtained layer is used for detection and localization. The final location coordinates and confidence are output after nonmaximum suppression. Experimental results on the Pascal VOC 2007 and 2012 datasets confirm that there is a 8.2% improvement in mAP compared to the original SSD and some existing algorithms.