Based on the annual average PM2.5 concentration data, economic data and health data from 2017 to 2020 in 2+26 cities, we evaluated the health effects and health and economic losses attributable to ${\rm PM}_{2.5}$ pollution in eight health end points (premature death, outpatient visit, hospitalization and illness). The results showed that the average annual ${\rm PM}_{2.5}$ concentration, health effects of each disease, economic loss, total health effects and total economic loss in 2+26 cities gradually decreased from 2017 to 2020, but the health effects and economic loss in some cities increased in 2020. The health endpoints with the highest economic losses were chronic bronchitis and premature death, followed by acute bronchitis, asthma, asthma, and asthma, followed by acute bronchitis, asthma, cardiovascular disease, and respiratory hospitalizations, and finally pediatrics and internal medicine. Based on the above results, the reduction o ${\rm PM}_{2.5}{\rm f}$ concentration was the main reason for the reduction of health effects and health economic effects attributed to ${\rm PM}_{2.5}$ in 2+26 cities. There is still room for further reduction of PM2.5 pollution in each city. Therefore, cities can still take corresponding measures to control ${\rm PM}_{2.5}$ pollution in the future to reduce health benefits and health economic benefits caused by ${\rm PM}_{2.5}.$ However, we also found that the reduction of health effects and health economic effects and the magnitude of reduction differed greatly among cities with different levels of development. Therefore, for some cities with a low level of economic development, they are facing the pressure of economic development and the pressure of reducing pollution. How to achieve the effect of reducing haze pollution while developing the economy is the problem that these cities need to solve at present.

Most existing models of RGB-D salient object detection (SOD) utilize heavy backbones like VGGs and ResNets which lead to large model size and high computational costs. In order to improve this problem, a lightweight two-stage decoder network is proposed. Firstly, the network utilizes MobileNet-V2 and a customized backbone to extract the features of RGB images and depth maps respectively. In order to mine and combine cross-modality information, cross reference module is used to fuse complementary information from different modalities. Subsequently, we design a feature enhancement module to enhance the clues of the fused features which has four parallel convolutions with different expansion rates. Finally, a two-stage decoder is used to predict the saliency maps, which processes high-level features and low-level features separately and then merges them. Experiments on 5 benchmark datasets comparing with 10 state-of-the-art models demonstrate that our model can achieve significant improvement with smallest model size.

Accurate image segmentation is an essential step in image processing. Gaussian mixture model (GMM) has been widely used for image segmentation due to its low complexity and high accuracy. However, the model assumes that the intensity distributions of images are symmetric, which makes it hard to obtain ideal results for images with asymmetric distributions. In addition, the model does not consider any noise, which makes it difficult to obtain ideal distribution fitting results when the image contains severe noises. Furthermore, the model only considers the distribution information without any spatial information, so it is sensitive to noise when segmenting images. To address these issues, we model noise with a Gaussian distribution and couple it into a skewed normal mixture model to reduce the effect of asymmetric distributions and noise and can obtain more accurate distribution fitting results. To further reduce the effect of noise, we propose a new anisotropic spatial information constraint term that preserves detailed information while reducing the effect of noise. Finally, an improved EM algorithm is proposed to solve the parameters of the model. Experimental results on synthetic and natural images show that our method achieves better segmentation results compared to other models.

Because of the excellent performance and fast speed of deep neural network, U-Net has become the most popular network framework for medical image segmentation. For various specific image segmentation tasks, researchers have proposed a series of U-Net related methods. However, on the one hand, due to the inherent limitations of convolutional neural networks, the variants of U-Net still cannot model long-range information well while maintaining detailed texture information. On the other hand, since medical images are difficult to obtain a large number of high-quality semantic pixel-level annotations, it is difficult to use supervised deep learning networks. To address these issues above, we proposed a modified U-Net structure and a Gaussian mixture model (GMM) based loss function. This modified U-Net can be well applied to brain MR image segmentation, which can not only restore the detailed information well, but also take into account the relatively large-scale local information. The proposed GMM loss can be used for unsupervised training of neural networks. It effectively alleviates the shortcomings of difficult access to medical image annotation data and improves the performance of deep neural networks. In the experiments in this paper, the GMM loss function can also be used as a regular term to assist supervised learning to achieve better results. Experimental results on brain MR images demonstrate the superior performance of the proposed model.

We propose a new Kantorovich theorem for Newton's method on Lie groups for mappings and matrix low-rank optimization problems, which arises from many applications. Under the classical hypothesis of f, we establish the convergence criteria of Newton's method from Lie group to its Lie algebra with weakened conditions, which improves the corresponding results in [20].