Target Recognition Of Underground Cavity Based On Time Series Classification Algorithm

Roads are important facilities in cities,and ensuring their safety is an important part of urban construction.Underground voids are one of the hidden dangers that threaten road safety.As a method of underground space exploration,ground penetrating radar plays an important role in the daily maintenance of roads.The ground penetrating radar receives the electromagnetic waves emitted by its own antenna,and image the signal through post-processing.At present,the widely used method of underground voids identification is based on the manual interpretation of the ground penetrating radar image,but the manual interpretation requires efficient operators.Therefore,the introduction of machine learning and deep learning is an inevitable trend in the recognition of underground voids by ground penetrating radar.In response to this trend,this study puts forward the idea of using the underground voids target recognition based on the time series classification algorithm.This study first analyzes the actual data set that has been collected.Due to the limited number of data sets,this article uses Gpr Max3.0 to generate simulated datasets as upplement to the actual datasets.In this paper,the two-dimensional image datasets and the three-dimensional image datasets are pre-processed and pre-screened based on energy detection respectively,and the corresponding parts of the underground voids are extracted from the images for further processing.Images contains a large amount of data,which are not appropriate for later training.In order to reduce the dimensionality of the data while retaining important information,this paper uses edge histogram descriptors,directional gradient histograms,and Log-Gabor filter.Image feature extraction is performed to obtain feature vectors,and preprocessing methods for feature extraction of three-dimensional images are designed.The feature vectors obtained by feature extraction are respectively input into Hidden Markov Model classifier,Long Short-Term Memory Neural Network and Gated Recurrent Unit Network for processing,and finally the recognition results are obtained.Based on the existing datasets,this paper gives the specific recognition effects of different methods,and discusses and analyzes the parameter settings and performance differences of various methods in two-dimensional image and three-dimensional image recognition.