Aiming at the problems of blurred target boundaries and low image segmentation efficiency in the image segmentation process of ancient murals, a multi-class image segmentation model fused with a lightweight convolutional neural network named MC-DM (Multi-Class DeepLabV3+MobileNetV2 (Mobile Networks Vision 2)) was proposed. In the model, DeepLabV3+ architecture and MobileNetV2 network were combined together, and the unique spatial pyramid structure of DeepLabV3+ was utilized to perform multi-scale fusion of the convolutional features of the mural to reduce the loss of image details during the mural segmentation. First of all, the features of the input image were extracted by MobileNetV2 to ensure the accurate extraction of image information and reduce the time consumption at the same time. Secondly, the image features were processed through the dilated convolution, so that the receptive field was expanded, and more semantic information was obtained without changing the number of parameters. Finally, the bilinear interpolation method was utilized to up-sample the output feature image to obtain a pixel-level prediction segmentation map, so that the accuracy of image segmentation was ensured to the greatest extent. In the JetBrains PyCharm Community Edition 2019 environment, a dataset made of 1 000 mural scanning pictures was used for testing. Experimental results showed that the MC-DM model had a 1% improvement in training accuracy compared with the traditional SegNet (Segment Network)-based image segmentation model, and had a 2% improvement in accuracy compared with the image segmentation model based on PSPNet (Pyramid Scene Parsing Network), and the Peak Signal-to-Noise Ratio (PSNR) of the MC-DM model was 3 to 8 dB higher than those of the experimental comparison models on average, which verified the effectiveness of the model in the field of mural segmentation. The proposed model provides a new idea for the segmentation of ancient mural images.

The scalar neurons of convolutional neural networks cannot express the feature location information,and have poor adaptability to the complex vehicle driving environment,resulting in low traffic sign recognition rate. Therefore,an intelligent traffic sign recognition method based on capsule network was proposed. Firstly,the very deep convolutional neural network was used to improve the feature extraction part. Then,a pooling layer was introduced in the main capsule layer. Finally,the movement index average method was used for improving the dynamic routing algorithm. The test results on the GTSRB dataset show that the improved capsule network method improves the recognition accuracy in special scenes by 10. 02 percentage points. Compared with the traditional convolutional neural network,the proposed method has the recognition time for single image decreased by 2. 09 ms. Experimental results show that the improved capsule network method can meet the requirement of accurate and real-time traffic sign recognition.

Aiming at the problem of low reconstruction quality of the existing Compressed Sensing Magnetic Resonance Imaging (CSMRI) algorithms at low sampling rates, an imaging method combining deep priors and non-local similarity was proposed. Firstly, a deep denoiser and Block Matching and 3D filtering (BM3D) denoiser were used to construct a sparse representation model that can fuse multiple priori knowledge of images. Secondly, the undersampled k-space data was used to construct a compressed sensing magnetic resonance imaging optimization model. Finally, an alternative optimization method was used to solve the constructed optimization problem. The proposed algorithm can not only use the deep priors through the deep denoiser, but also use the non-local similarity of the image through the BM3D denoiser to reconstruct the image. Compared with the reconstruction algorithms based on BM3D, experimental results show that the proposed algorithm has the average peak signal-to-noise ratio of reconstruction increased about 1 dB at the sampling rates of 0.02, 0.06, 0.09 and 0.13. Compared with the existing MRI algorithm WaTMRI (Magnetic Resonance Imaging with Wavelet Tree sparsity),DLMRI (Dictionary Learning for Magnetic Resonance Imaging), DUMRI-BM3D (Magnetic Resonance Imaging based on Dictionary Updating and Block Matching and 3D filtering), etc, the images reconstructed by the proposed algorithm contain a lot of texture information, which are the closest to the original images.

To improve the performance of vehicle classification by making full use of image information from traffic surveillance, Histogram of Oriented Gradient Convolutional (HOG-C) features extraction method was added on the capsule network, and a Capsule Network model fusing with HOG-C features (HOG-C CapsNet) was proposed. Firstly, the gradient data in the images were calculated by the gradient statistical feature extraction layer, and then the Histogram of Oriented Gradient (HOG) feature map was plotted. Secondly, the color information of the image was extracted by the convolutional layer, and then the HOG-C feature map was plotted with the extracted color information and HOG feature map. Finally, the HOG feature map was input into to the convolutional layer extract its abstract features, and the abstract features were encapsulated through a capsule network into capsules with the three-dimensional spatial feature representation, so as to realize the vehicle classification by dynamic routing algorithm. Compared with other related models on the BIT-Vehicle dataset, the proposed model has the accuracy of 98.17%, the Mean Average Precision (MAP) of 97.98%, the Mean Average Recall (MAR) of 98.42% and the comprehensive evaluation index of 98.20%. Experimental results show that the vehicle classification in traffic surveillance scenarios can be achieved with better performance by using HOG-C CapsNet.

Ranking function based annotation may cause more or fewer labels according to the probability predicted by the model in multi-label image annotation. Therefore, a Convolutional Neural Network with THreshold OPtimization (CNN-THOP) model was proposed. The model consists of Convolutional Neural Network (CNN) and threshold optimization. Firstly, CNN was used to train a model, which was used to predict the image, so as to obtain the prediction probability, and Batch Normalization (BN) layer was added to the CNN to effectively accelerate the convergence. Secondly, threshold optimization was performed by the prediction probabilities of the test set images obtained by the proposed model. After the threshold optimization process, an optimal threshold was obtained for each kind of label, so as to obtain a set of optimal thresholds. Only when the prediction probability of this kind of label was greater than or equal to the best threshold of this kind of label, the image would be labeled with this label. In the labeling process, the CNN model and a set of optimal thresholds were added to achieve more flexible multi-label labeling of the image to be labeled. Through the verification on 8 000 images in the natural scene image dataset, experimental results show that CNN-THOP has about 20 percentage points improvement on average precision compared to Ranking Support Vector Machine (Rank-SVM), and is about 6 percentage points and 4 percentage points higher respectively than Convolutional Neural Network using Mean Square Error function (CNN-MSE) in average recall and F1 value respectively, and has the Complete Matching Degree (CMD) reached 64.75%, which proves that the proposed method is effective in automatic image annotation.