The increasingly frequent marine oil spill disasters have great harm to the marine ecosystem. As an essential means of remote sensing monitoring, synthetic aperture radar (SAR) images can detect oil spills in time and reduce marine pollution. Many look-alike oil spill regions are difficult to distinguish in SAR images, and the scarcity of real oil spill data makes it difficult for deep learning networks to train effectively. In order to solve the above problems, this paper designs a multi-task generative adversarial networks (MTGANs) oil spill detection model to distinguish oil spills and look-alike oil spills and segment oil spill areas in one framework. The discriminator of the first GAN is transformed into a classifier, which can effectively distinguish between real and look-alike oil spills. The generator of the second GAN model integrates a fully convolutional symmetric structure and multiple convolution blocks. Multiple convolution blocks can extract the shallow oil spill information, and the fully convolutional symmetric structure can extract the deeper features of the oil spill information. The algorithm only needs to use a small number of oil spill images as the training set to train the network, and the limitation of the oil spill dataset can be solved. Validation evaluations are conducted on three datasets of Sentinel-1, ERS-1/2 and GF-3 satellites, and the experimental results demonstrate that the proposed MTGANs oil spill detection framework outperforms other models in oil spill classification and semantic segmentation. Among them, the classification accuracy of the oil spill and look-alikes can reach 97.22 $\%$ . The average OA for semantic segmentation of the oil spill area can be 97.47 $\%$ and the average precision can reach 86.69 $\%$ . The code of this work will be available at https://github.com/fjc1575/Marine-Oil-Spill/tree/main/MTGANs for the sake of reproducibility.

Offshore aquaculture raft information extraction from synthetic aperture radar (SAR) images is essential for large-scale marine resource exploitation and protection. In this letter, a deep learning model called multi-scaled attention U-net with dilated convolution and offset convolution (MDOAU-net) is proposed for aquaculture raft monitoring via SAR image segmentation. The U-net backbone and attention gate of the Attention U-net are used in the MDOAU-net model. In addition, the MDOAU-net model consists of three distinctive parts. First, a multi-scale feature-fusion block is adopted in its input to extract features from raw images. Moreover, adapted from the Attention U-net for SAR image segmentation, fewer channels are used in each convolution layer of the MDOAU-net to match latent features in SAR images. Furthermore, nine dilated convolution blocks are adopted in the encoder–decoder structure to extract semantic features in the presence of speckle noises. In addition, offset convolution blocks are developed to convert spatial information into channel information for the precise segmentation of blurry boundaries. Four skip connections of the U-net backbone are replaced by four offset convolution blocks. Experimental results are elaborated to demonstrate the superior performance of the MDOAU-net model to seven existing methods in terms of overall accuracy (OA) and number of parameters.

Land cover classification using polarimetric synthetic aperture radar (PolSAR) images is an important tool for remote sensing analysis. In view that PolSAR image effective interpretation is commonly affected by the absence of discriminative features and the presence of speckle noises, this letter proposes an adaptive nonlocal stacked sparse autoencoder (ANSSAE) to achieve PolSAR image classification. It extracts the adaptive nonlocal spatial information by adaptively calculating weighted average values of each pixel from nonlocal regions, which can reduce the influence of speckle noises and retain edge details. In the first layer of the ANSSAE, the adaptive nonlocal spatial information is introduced into the objective function to obtain the robust feature representation, whose effects would transfer to the rest of layers. Therefore, the ANSSAE can automatically capture spatial-related, robust, and distinguishable features, which can suppress speckle noises and gain accurate classification results. Experimental results on two real PolSAR images demonstrate that the proposed approach can significantly improve the classification accuracy.

Mangroves are an important source of blue carbon that grow in coastal areas. The study of mangrove species distribution is the basis of carbon storage research. In this study, we explored the potential of combining optical (Gaofen-1, Sentinel-2, and Landsat-9) and fully polarized synthetic aperture radar data from different periods (Gaofen-3) to distinguish mangrove species in the Fucheng town of Leizhou, Guangdong Province. The Gaofen-1 data were fused with Sentinel-2 and Landsat-9 satellite data, respectively. The new data after fusion had both high spatial and spectral resolution. The backscattering coefficient and polarization decomposition parameters of the fully polarized SAR data which could characterize the canopy structure of mangroves were extracted. Ten different feature combinations were designed by combining the two types of data. The extremely randomized trees algorithm (ERT) was used to classify the species, and the optimal feature subset was selected by the feature selection algorithm on the basis of the ERT, and the importance of the features was sorted. Studies show the following: (1) When controlling a single variable, the higher the spatial resolution of the multi-spectral data, the higher the interspecific classification accuracy. (2) The coupled Sentinel-2 and Landsat-9 data with a 2 m resolution will have higher classification accuracy than a single data source. (3) The selected feature subset contains all types of features in the optical data and the polarization decomposition features of the SAR data from different periods: multi-spectral band > texture feature > polarization decomposition parameter > vegetation index. Among the optimized feature combinations, the classification accuracy of mangrove species was the highest, the overall classification accuracy was 90.13%, and Kappa was 0.84, indicating that multi-source and SAR data from different periods coupling could improve the discrimination of mangrove species. (4) The ERT classification algorithm is suitable for the study of mangrove species classification, and the classification accuracy of extremely random trees in this paper is higher than that of random forest (RF), K-nearest neighbor (KNN), and Bayesian (Bayes). The results can provide technical guidance and data support for mangrove species monitoring based on multi-source satellite data.

Marine aquaculture is an important natural resource exploration that requires rational planning to avoid environmental damage. Synthetic aperture radar (SAR) images are essential in remote sensing to monitor the marine ecological environment. Unsupervised methods provide an adequate solution to avoid the cost of training sample collection. However, unsupervised methods often struggle to discover effective semantic information and incrementally use newly acquired data. To address these challenges, this article presents an incremental double unsupervised deep learning (IDUDL) model, which is specially designed to characterize unlabeled marine aquaculture and achieve the results semantically. Based on the idea of alternately generating and updating pseudo-labels, the proposed IDUDL model defines the double neural networks comprising the feature extraction network (FEN) and the fully convolutional semantic segmentation network (FCSSN). A patch estimation (PE) is proposed to generate pseudo-labels with aquaculture semantic information based on the features extracted by the FEN. Then, the aquaculture extraction results are obtained by FCSSN with generated pseudo-labels. After that, the pseudo-labels and extraction results are updated in turn until the pseudo-labels are stable. In addition, due to the unique structure of double neural networks, newly acquired marine aquaculture SAR images can also be added to the pretrained FCSSN and followed pseudo-labels updated based on the FEN and PE part, which can achieve new data incremental learning without retraining the whole IDUDL model. Experiments demonstrate the effectiveness of the proposed approach based on two different ways of marine aquaculture including raft and cage types, which consist of GaoFen-3 (GF-3) and RADARSAT-2 SAR images from the Dalian and Ningde areas, respectively. The incremental experiments are also designed to verify the generalization of the IDUDL model for newly obtained marine aquaculture SAR images. The code of this work will be available at https://github.com/fjc1575/Marine-Aquaculture/tree/main/IDUDL for the sake of reproducibility.