The current limited spaceborne hardware resources and the diversity of ship target scales in SAR images have led to the requirement of on-orbit real-time detection of ship targets in spaceborne synthetic aperture radar (SAR) images. In this paper, we propose a lightweight ship detection network based on the YOLOv4-LITE model. In order to facilitate the network migration to the satellite, the method uses MobileNetv2 as the backbone feature extraction network of the model. To solve the problem of ship target scale diversity in SAR images, an improved receptive field block (RFB) structure is introduced, enhancing the feature extraction ability of the network, and improving the accuracy of multi-scale ship target detection. A sliding window block method is designed to detect the whole SAR image, which can solve the problem of image input. Experiments on the SAR ship dataset SSDD show that the detection speed of the improved lightweight network could reach up to 47.16 FPS, with the mean average precision (mAP) of 95.03%, and the model size is only 49.34 M, which demonstrates that the proposed network can accurately and quickly detect ship targets. The proposed network model can provide a reference for constructing a spaceborne real-time lightweight ship detection network, which can balance the detection accuracy and speed of the network.

The shipping industry is pivotal in transporting approximately 90% of the world’s goods, and it is characterized by evolving trends in vessel sizes and energy-efficient designs. Continuous advancements in technology for ship management have focused on detecting and analyzing anomalous and illicit vessels. In this study, we introduce ShipGeoNet, a model designed to extract geometric features from ships captured in Sentinel-1 synthetic aperture radar (SAR) images. ShipGeoNet employs a combination of convolutional neural networks (CNNs) and nonlinear regression techniques to extract various geometric features of ships from SAR imagery. The model follows a two-step approach. First, it utilizes a modified Mask R-CNN architecture and the ViTDet model to accurately detect ships, generating high-quality object masks for precise localization. In the subsequent step, a regression model utilizes the detected ship masks to extract key geometric attributes, including length, width, and orientation. The proposed nonlinear regression techniques are specifically crafted to address the complex nonlinear deformations inherent in SAR images. Through extensive experiments on a large-scale SAR dataset, ShipGeoNet demonstrates its efficiency and accuracy in ship size extraction and matching, outperforming existing methods. Developing the ShipGeoNet model opens up possibilities for future applications in maritime surveillance, navigation, and environmental monitoring.

Autoencoders (AEs) are commonly utilized for acquiring low-dimensional data representations and performing data reconstruction, which makes them suitable for hyperspectral unmixing. However, AE networks trained pixel by pixel and those employing localized convolutional filters disregard the global material distribution and distant interdependencies, resulting in the loss of necessary spatial feature information essential for the unmixing process. To overcome this limitation, we propose an innovative deep neural network model named U-shaped transformer network using shifted windows (UST-Net). UST-Net prioritizes spatial information in the scene that is more discriminative and significant by using multi-head self-attention blocks based on shifted windows. Unlike patch-based unmixing networks, UST-Net operates on the complete image, eliminating inconsistencies associated with patches. Moreover, the downsampling and upsampling stages are used to extract HSI feature maps at different scales. This process generates a context-rich and spatially accurate abundance map without losing local details. The experimental results of one synthetic dataset and three real datasets demonstrate that UST-Net significantly outperforms both traditional and several other advanced neural network methods. Our code is publicly available at https://github.com/UPCGIT/UST-Net.

With a gradually rising global average sea level, it is of great significance to predict changes in sea level. However, sea level variations often exhibit both linear and nonlinear characteristics, complicating the prediction of sea level changes with a single model. The seasonal autoregressive integrated moving average (SARIMA) model can fully consider the linear characteristics of time series, but its nonlinear prediction ability is poor; the long short-term memory (LSTM) model can compensate for this shortcoming. To predict complex sea level changes, we propose a strategy to combine the SARIMA and LSTM models to increase the sea level prediction accuracy. In our method, sea level anomaly (SLA) time series are decomposed into the trend and seasonal term and random term; then, the SARIMA model is used to predict the trend and seasonal term of sea level variations, while the random term is predicted by LSTM. Sea surface height data from 1993 to 2018 are used in an experiment. Compared with other models, the performance of the SARIMA+LSTM model is superior in predicting sea level changes with a minimum root mean square error of 1.155 cm and a maximum determination coefficient (R2) of 0.89 during the testing period. Furthermore, the predicted results are in close agreement with the SLA data, which indicates that the SARIMA+LSTM model could be successfully used for the estimation of sea level variability.

Oil spill pollution at sea causes significant damage to marine ecosystems. Quad-polarimetric Synthetic Aperture Radar (SAR) has become an essential technology since it can provide polarization features for marine oil spill detection. Using deep learning models based on polarimetric features, oil spill detection can be achieved. However, there is insufficient feature extraction due to model depth, small reception field lend due to loss of target information, and fixed hyperparameter for models. The effect of oil spill detection is still incomplete or misclassified. To solve the above problems, we propose an improved deep learning model named BO-DRNet. The model can obtain a more sufficiently and fuller feature by ResNet-18 as the backbone in encoder of DeepLabv3+, and Bayesian Optimization (BO) was used to optimize the model’s hyperparameters. Experiments were conducted based on ten prominent polarimetric features were extracted from three quad-polarimetric SAR images obtained by RADARSAT-2. Experimental results show that compared with other deep learning models, BO-DRNet performs best with a mean accuracy of 74.69% and a mean dice of 0.8551. This paper provides a valuable tool to manage upcoming disasters effectively.