Concerning the problems of poor classification performance and high computational overhead of Capsule Network （CapsNet） on complex images with background noise information， an improved capsule network model based on attention mechanism and weight sharing was proposed， called Shared Transformation Matrix CapsNet （STM-CapsNet）. The proposed model mainly includes the following improvement. 1） An attention module was introduced into the feature extraction layer of CapsNet， which enabled low-level capsules to focus on entity features related to the classification task. 2） Low-level capsules with close spatial positions were divided into several groups， and each group of low-level capsules was mapped to high-level capsules by sharing transformation matrices， which reduced computational overhead and improved model robustness. 3） The L₂ regularization term was added to margin loss and reconstruction loss to prevent model overfitting. Experimental results on three complex image datasets including CIFAR10， SVHN （Street View House Number） and FashionMNIST show that， the above improvements are effective in enhacing the model performance； when the number of iterations is 3， and the number of shared transformation matrices is 5， the average accuracies of STM-CapsNet are 85.26%， 93.17% and 94.96% respectively， the average parameter amount is 8.29 MB， verifying that STM-CapsNet has better performance compared with the baseline models.

The current real-time semantic segmentation algorithm has the high computational cost and large memory footprint， which cannot meet the applications requirements of actual scenes. In order to solve the problems， a new type of shallow lightweight real-time semantic segmentation algorithm — AEFNet （Real-time segmentation algorithm based on Attention mechanism and Effective Factorized convolution） was proposed. Firstly， one-dimensional non-bottleneck structure （Non-bottleneck-1D） was adopted to construct a lightweight factorized convolution module to extract rich contextual information and reduce the amount of calculation. At the same time， the learning ability of the algorithm was enhanced in a simple way and the extraction of detailed information was facilitated. Then， the pooling operation and Attention Refinement Module （ARM） were combined to construct a global context attention module to capture global information and refine each stage of the algorithm to optimize the segmentation effect. The algorithm was verified on the public datasets cityscapes and camvid， and the precision of 74.0% and the inference speed of 118.9 Frames Per Second （FPS） were obtained on the cityscapes test set. Compared with Depth-wise Asymmetric Bottleneck Network （DABNet）， the proposed algorithm has the precision increased by about 4 percentage points， and the inference speed increased by 14.7 FPS. Compared with the recent efficient Enhanced Asymmetric Convolution Network （EACNet）， the proposed algorithm has the precision slightly lower by 0.2 percentage points， but has the inference speed increased by 6.9 FPS. Experimental results show that the proposed algorithm can more accurately identify the scene information， and can meet the real-time requirements.