The Study Of Low-level Wind Shear Recognition Based On Multi-feature Fusion

Low-level wind shear has become a dangerous factor that seriously influenced the aircraft flight safety, and it has different effects to the aircraft flight along with the variation of types of wind shear because of the unique wind field characteristics. In order to ensure flight safety, the pilot needs to respond differently to the different types of wind shear. So it is specifically important to recognize the different types of wind shear correctly.Because of the lack of real lidar data of wind shear, the research of type recognition of four typical types of wind shear which are microburst, low-level jet, sea breeze and crosswind is made in this paper using the simulation three-dimensional wind field. The specific work is as follows:Firstly, on the condition that ignoring the factors of specific terrain and climate,constructing the simulation models of four typical types of wind shear which are microburst,low-level jet, sea breeze and crosswind through using atmospheric fluid software FLUENT based on Computational Fluid Dynamic, and on the condition of different elevations, distance and azimuths, simulating PPI scanning method of Doppler lidar to scan wind field. Then transforming the acquired scanning data to two-dimensional image and constructing sample book.Secondly, for the characteristics of the images of sample book, a low-level wind shear type recognition method which is based on shape and texture features is proposed in this paper. Firstly, shape features which reflect the global changes of wind field and texture features which reflect the local changes of wind field are extracted by Zernike moments and rotation invariant uniform Local Binary Pattern(LBP). Then, the two features are combined in series and the combined features are reduced by Principal Component Analysis(PCA) in order to get the effective features. Finally, k-nearest neighbor classifier is used to classify four types of low-level wind shear images. The experimental results show that the proposed algorithm has better recognition performance which is improved significantly compared with the single algorithm above.Thirdly, in the process of single Doppler lidar scanning, it can scan a elevation every time. But the low-level wind shear occurs unexpectedly and the duration is short, single elevation scanning may arise problems of missing or incomplete scanning. So this paper proposes a scanning method that is using several lidar to scanning the wind field with multielevation at the same time to acquire multi-level wind field feature, and doing the type recognition of low-level wind shear. The experimental results show that compared with the recognition single elevation scanning image this method has acquired certain improvement.