| 3D seismic exploration has a high demand for seismic data acquisition accuracy.The exploration area covers a wide range of areas and the distribution of obstacles in the area is complex,which directly affects the seismic data acquisition of the three-dimensional observation system.Therefore,variable design can be divided into two ways:artificial and automatic variable design.Artificial observation method is mainly used for obstacle avoidance processing through manual obstacle labeling of remote sensing data and actual obstacle materials collected in the work area.However,with the development of three-dimensional seismic acquisition,exploration work area is becoming more and more complex.Traditional obstacle avoidance methods are not only inefficient but also unreliable.In recent years,in-depth learning has achieved great success in the field of image recognition.In order to use remote sensing satellite images,the traditional obstacle avoidance methods are becoming Automatic recognition of obstacles provides a good solution.Based on remote sensing satellite images,obstacles in exploration area are automatically identified,and shot points are automatically optimized according to evaluation criteria,which makes the variable design more meet the production needs of three-dimensional seismic exploration and more intelligent.Firstly,the method of automatic obstacle recognition is studied.Threshold segmentation,K-means clustering,Support Vector Machine(SVM)and Deep Learning(Deep Learning)algorithms are used to automatically recognize obstacles in remote sensing satellite images.Threshold segmentation and K-means clustering belong to unsupervised algorithm,which can recognize and segment obstacles directly without prior information of training samples for model training.Support Vector Machine(SVM)and in-depth learning belong to supervised algorithms,which need to use training samples for model training,in which support vector machine can select small samples to train obstacles in remote sensing images,which can achieve better recognition accuracy;in-depth learning uses open remote sensing data sets for model training,and then uses remote sensing images of actual work areas for migration learning to realize obstacle recognition.Don’t split up.Through comparative analysis,it is found that the recognition accuracy of supervised algorithm is much higher than that of unsupervised algorithm when there are more training samples.After the obstacle recognition is completed,the global optimization method is further used to design the shot point.The result of obstacle recognition is used to pick up contour boundary automatically.The heuristic simulated annealing algorithm is used to optimize the location of shot points with the minimum mean square deviation of coverage times as the convergence condition.After analyzing the three-dimensional observation system before and after the change of view,the coverage times of the observation system are improved obviously,and the purpose of the change of view is achieved. |
