Influenced by the microscopic sensors and irregular collection method, the pollen images are often disturbed by different degrees of noise and have rotation changes with different angles, which leads to generally low recognition accuracy. In order to solve the problem, a Dominant Gradient encoding based Local Binary Pattern (DGLBP) descriptor was proposed and applied to the recognition of pollen images. Firstly, the gradient magnitude of an image block in the dominant gradient direction was calculated. Secondly, the radial, angular and multiple gradient differences of the image block were calculated separately. Then, the binary coding was performed according to the gradient differences of each image block. The binary coding was assigned weights adaptively with reference to the texture distribution of each local region, and the texture feature histograms of pollen images in three directions were extracted. Finally, the texture feature histograms under different scales were fused, and the Euclidean distance was used to measure the similarity between images. The average correct recognition rates of DGLBP on datasets of Confocal and Pollenmonitor are 94.33% and 92.02% respectively, which are 8.9 percentage points and 8.6 percentage points higher on average than those of other compared pollen recognition methods, 18 percentage points and 18.5 percentage points higher on average than those of other improved LBP-based methods. The experimental results show that the proposed DGLBP descriptor is robust to noise and rotation change of pollen images, and has a better recognition effect.

In order to improve the accuracy of haze forecast and resolve the time lagging and low accuracy of temporal model, a mixed forecast method based on time series analysis and Karman filter was proposed. Firstly, the stability of time series was tested by graph analysis and eigenvalue analysis (ADF). Unstable time series were converted to stable ones by differential operation. A statistical function was established based on the stable time series. And then, the obtained model equations were used as the state and observation equation for Kalman filtering. Final haze forecast was based on recursion by Karman filtering. The experimental results showed that the accuracy of haze forecast is effectively improved by the mixed forecast method based on time series analysis and Karman filtering.

According to the problem that the existing roughness descriptors are mostly dependent on the average grey value, which is easy to cause the loss of image information, a new roughness descriptor based on Gaussian scale space was presented for pollen image classification and recognition. With this method, the Gaussian pyramid algorithm was used to divide the image into several different levels of scale space, and then the roughness texture feature was extracted from the different level scale space. The statistical distribution of roughness frequency was calculated to build the Scale-Space Roughness Histogram Descriptor (SSRHD). At last, the Euclidean distance was used to measure the similarity between images. The simulation results on Confocal and Pollenmonitor image database demonstrate that, compared with Discrete Hidden Markov Model Descriptors (DHMMD), the Correct Recognition Rate (CRR) performed by the SSRHD increases by 2.32% on Confocal and 1.2% on Pollenmonitor, and the False Recognition Rate (FRR) decreases by 0.1% on Confocal. The experimental results show that the SSRHD feature can effectively describe the pollen image texture and it also has good robustness to pollen rotation and pose variation.

The 3D virtualization of cloud data has always been the hotspot of computational graphics and meteorology. A method for data modeling and rendering based on Weather Research and Forecasting (WRF) was proposed to realize the 3D virtual simulation of real-world cloud data. Due to the complexity of particle system modeling and its poor real-time performance, a WRF cloud model was first set up, regarding the relationship between particles in the cloud system; further illumination rendering and 3D simulation were completed based on the illumination model and billboard technique; meanwhile, Imposter technique was introduced to speed up the texture mapping and improve the performance. The simulation results show that the proposed method owns the benefits of fast modeling and rendering of cloud data as well as good fidelity of the 3D virtualization model.