| With the continuous advancement of science and technology in society,people's production and lifestyles are gradually developing in the direction of intelligence and automation.In the field of meteorology and aviation,with the continuous introduction of deep learning technology,how to use deep learning to automate such professional problems as cloud cluster identification,cloud cluster classification,cloud cluster segmentation,aircraft automated navigation,and navigation route planning has always been Hot issues studied by related researchers.The satellite cloud image contains a wealth of information,and the various cloud information contained therein is the indispensable basic data for weather forecasting and route planning in the fields of meteorology and aviation.In view of the huge satellite cloud image data and the limitations of using manual extraction and analysis methods,the use of artificial intelligence for data processing is particularly important.At this stage,meteorological information about thunderstorms and other weather information is confirmed and identified by meteorological staff.This manual identification method cannot meet the real-time requirements for weather warning,and brings great challenges to people's personal and property safety protection.In order to solve the above problems,this paper proposes a network structure that combines separable convolution and residual blocks,and uses a multi-pool structure to extract global information.Plan the flight trajectory and complete the auxiliary decision-making task.The main research contents of this article include:1)On the basis of reading a large number of documents,the necessity of intelligent identification of satellite cloud images is analyzed,and the current research status of image segmentation and thunderstorm cloud cluster identification in satellite cloud images at home and abroad are summarized,and the commonly used ones at this stage are analyzed.Segmentation methods and the limitations of their use in satellite cloud images.2)Introduced the basic principles and hierarchical structure of convolutional neural network in detail.Aiming at the problem that the parameters of ordinary convolution are too large,the calculation efficiency is low,and it cannot be directly used in embedded devices,the convolutional neural network is carried out.Lightweight improvements.A method of combining separable convolution and residual block is proposed,which reduces the amount of parameters in the network and improves the computational efficiency of the network.The back-end of the network uses pooling windows of different sizes to extract detailed information in the feature map,which improves The recognition accuracy of thunderstorm cloud clusters.Finally,the segmentation results of thunderstorm cloud clusters are compared with commonly used image segmentation methods.The network structure in the article has advantages over commonly used network segmentation networks,and the recognition accuracy of thunderstorm cloud clusters has increased by 0.5%,and it is better for satellite cloud images.Response time has been reduced by 10%.It proves that the relevant network structure can identify the thunderstorm cloud cluster in the satellite cloud image,and the response time is reduced.3)Based on the satellite cloud image data that has been identified for thunderstorm cloud clusters,the flight path of the aircraft is planned using the depth-first algorithm with flags and gradient vectors added,which reduces the calculation loss of the depth-first algorithm and improves the calculation efficiency of the algorithm.Finally,experiments prove that the improved DFS method can quickly and effectively plan routes,bypass thunderstorm clouds,and achieve the role of assisting flight route decision-making in the aviation field.At the end of the full text,in order to facilitate the use of some non-related personnel,the data analysis and network structure have been practically and systematically encapsulated. |
PDF Full Text Request