Obstacle Recognition Based On Feature Vectors In Flow Fields

In the research of obstacle identification in the flow field(fish side line bionics),there is a common problem that the flow field is too simple,such as the use of potential flow or periodic flow field.Compared with the majority of studies,a more complicated unsteady flow field with no obvious period is used in this study,which is generated by the interference of two cylinders,and applies the idea of ‘feature extraction – classifier’to the identification of obstacles in the flow field,and obtains good results.The method is as follows: After introducing the simulation parameters and verifying the correctness of the simulation,the original data of sampling points in each obstacle working condition is obtained by simulation.The eigenvalues representing the behavior of sampling points at all times were extracted and screened,and the original data was reduced to 8-dimensional eigenvectors,and the data set of each obstacle and working condition was constructed.Then the data set was used to train the three classifiers,namely neural network,SVM and linear discrimination.After the training is completed,the input of the model is the velocity sequence measured by a single sampling point at any position and any time in the sampling area of the flow field,and the output of the model is the obstacle and the type of working condition.Then,the predicted obstacle type is compared with the real one,and the specific confusion matrix is obtained.The obstacle identification task in the flow field is completed,and the performance of the model is measured by the accuracy.This study also investigated the influence of different sampling duration,sampling area and sampling point density on the recognition results,and investigated the anti-noise performance of the recognition model by adding Gaussian white noise with different signal-to-noise ratios to the original data.