Research On Classification Of Land And Sea Targets Based On Deep Learning

With the continuous development of emergency rescue,buildings and ships are the two most important targets in land and sea,and the classification of building earthquake damage and ship target classification plays an important role in emergency management.In earthquake disasters,rapid and accurate assessment of building damage can help rescuers make targeted rescue plans,improve rescue efficiency and success rate;and sea ship target classification can help emergency management departments monitor maritime traffic conditions,timely detection of marine traffic accidents,and improve the speed and level of emergency response.In earthquake disasters,buildings are one of the most vulnerable ground features,and their damage is directly related to the progress and work efficiency of disaster relief.The traditional manual survey method is limited by factors such as manpower,time and location,and it is difficult to quickly and accurately obtain the damage of buildings in the disaster area.The earthquake damage classification of buildings based on deep learning can quickly provide detailed disaster information and provide strong support for postdisaster rescue and reconstruction.Although many methods have been proposed in the study of seismic damage assessment of buildings,most of them divide damaged buildings into two categories—intact and damaged—which is in-sufficient to meet practical needs.To address this issue,this study presents a novel convolutional neural network—namely,the earthquake building damage classification net(EBDC-Net)—for assessment of building damage.The proposed network comprises a feature building damage feature extraction encoder module and a damage classification module.The feature extraction encoder module is employed to extract semantic information on building damage,and introduce a spatial attention mechanism to gather scattered damage features in the image,enhance the feature expression ability of network,and improve the ability to distinguish between different damage levels.While the classification module combines the global and contextual feature extraction modules to enhance the network’s ability to explore the relationship between global features and local features,and improves the accuracy of the network in classifying building damage levels.The performance of EBDC-Net was evaluated on the Ludian and Yushu earthquake building damage datasets,and a largescale damage assessment was performed on the Yangbi earthquake dataset.The results of the experiments indicate that this approach can quickly and accurately classify buildings with different damage levels.The overall classification accuracy was 94.44%,85.53%,and 77.49% when the damage to the buildings was divided into two,three,and four categories,respectively.In the maritime field,ship target classification is an important basis for maritime safety,resource development and environmental detection.Although various ship classification methods have been proposed during the past years,most of them are developed based on single source data,which is not enough to meet the needs of all-time and all-weather ship classification.At the same time,only use single data source will also affect the accuracy of ship classification.This study proposes an adaptive weighted decision-level fusion ship classification method based on multi-source data to solve the problem that only a single data source is used and the accuracy of ship classification is not high under complex sea conditions.The method includes a multi-source image generation module,a single-source data ship classification module and a multi-source data decision-level fusion classification module.Aiming at the problem of low accuracy of ship recognition using single source data,this paper proposes a novel ship recognition method using multi-source remote sensing images based on improved YOLOv4 and adaptive weighted decision level fusion.The multi-source image generation module builds an OPT-SAR image translation model based on the Cycle GAN network,translate optical images into matching SAR images,and establishes a multi-source remote sensing image ship classification dataset.The single-source data ship classification module introduces a Gaussian model to model the coordinates of the prediction box in the YOLOv4 model to improve the reliability of the prediction box.And identify different types of ships from SAR and optical images respectively.In the multi-source data decision-level fusion classification module,an adaptive weighted decision-level fusion classification method is constructed to perform decision-level fusion judgment on the classification results of different data sources,and improve the accuracy of ship classification under complex sea conditions.Compared with the state-of-the-art decisionlevel fusion classification methods;the proposed method in this study obtains the best accuracy;compared with only the SAR and optical images are used,the accuracy is improved 10.57% and 2.61%,respectively;Even under complex sea conditions,the method proposed still shows the good ability for ship classification.