Fig 1 - uploaded by Haitao Lang
Content may be subject to copyright.
Typical samples of challenges for both ship detection and category recognition in high-resolution synthetic aperture radar (SAR) imagery.
Source publication
Ship surveillance by remote sensing technology has become a valuable tool for protecting marine environments. In recent years, the successful launch of advanced synthetic aperture radar (SAR) sensors that have high resolution and multipolarimetric modes has enabled researchers to use SAR imagery for not only ship detection but also ship category re...
Contexts in source publication
Context 1
... respect to maritime security and environmental protection, ship surveillance by remote sensing has proven to be a key technology that has aided many maritime applications. 1 However, sensing technologies are often problematic, such as coastal radar that has limited coverage, opti- cal sensors that can only be used during daylight hours and low cloud cover, and automatic identification systems or vessel monitoring systems that require cooperation from ship crews. By contrast, synthetic aperture radar (SAR) is a powerful surveillance tool that allows for the collection of observations across broad expanses of open water, independently from the effects of the weather and from the day and night solar cycles. 2 The successful use of SAR imaging to detect and recognize ships depends on a variety of factors including ship size, sea conditions, and radar characteristics (e.g., polarization and inci- dence angle). These inherent issues make the detection and recognition of ships via SAR imagery very challenging. Early SAR sensors can only provide low-resolution images that amount to dozens of meters per pixel in both the azimuth and range directions. In such cases, ships appear as several bright points against sea clutters and the detection of ships often relies on a global or local threshold method, such as the constant false alarm rate (CFAR). 3,4 With these sensors, it is almost impossible to recognize the ship category. Advances in high-resolution SAR imaging technology (e.g., capable of achieving meter- level resolution for spaceborne SAR) have resulted in not only the acquisition of truer informa- tion (e.g., more structural information that makes it possible to recognize ship types) but also the presentation of new challenges for researchers who would like to develop methods for automatic ship detection and recognition via SAR imagery. Some typical challenges are shown in Fig. 1 and are described ...
Context 2
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Context 3
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Context 4
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Context 5
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Context 6
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Context 7
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Context 8
... Some sea clutter and the ship targets have approximative backscattering intensity in high-resolution SAR imagery [ Fig. 1(a)]. 2. The evident sidelobes submerge the structures of ships [ Fig. 1(b)]. 3. Ambiguities in SAR imagery are often mistaken as ship targets [ Fig. 1(c)]. 4. The similarity of detected ship appearances from different categories and the dissimi- larity of ships from the same category make it hard to classify ships correctly. For exam- ple, in Fig. 1, (d) and (f) are container ships and (e) is a cargo ship. However, the appearance of the ship in Fig. 1(d) is more like that in Fig. 1(e) instead of the ship in Fig. 1(f). Even for experts, it is a nontrivial challenge to distinguish them. 5. Although high-resolution SAR imagery provides the possibility to recognize ship cat- egory based on the backscattering pattern, it is still not a trivial task. Many low-level and mid-level image features that have been widely used in object recognition and scene classification applications cannot be introduced directly into ship category classifications via SAR imagery. For example, we can extract some low-level image features such as scale-invariant feature transform (SIFT) 5 and histogram of oriented gradient (HOG), 6 can train a classifier based on bag-of-word image representations 7 of Figs. 1(d′), 1(e′), and 1(f′), and then can categorize them into container and cargo classes correctly. However, when we use the same approach with SAR images [Figs. 1(d), 1(e), and 1(f)], we find that it is not effective (as shown in our experimental ...
Similar publications
Synthetic aperture radar (SAR) in Terahetz (THz) band is very important and valuable in the field of radar signal processing, and it is much sensitive to the high-frequency vibration of the platform due to the short wavelength. In this paper, the high-frequency vibration is characterized as multi-component sinusoidal frequency modulated (SFM) signa...
This paper proposes a 1-bit Synthetic Aperture Radar (SAR) imaging method based on a single-frequency time-varying threshold. Synthetic aperture radar echoes are quantized to 1-bit sampling data by comparing the data with the threshold; this reduces the data-width of the SAR echoes, consequently simplifying the system and improving efficiency. The...
Moving ship targets appear blurred and defocused in synthetic aperture radar (SAR) images due to the translation motion during the coherent processing. Motion compensation is required for refocusing moving ship targets in SAR scenes. A novel refocusing method for moving ship is developed in this paper. The method is exploiting inverse synthetic ape...
Moving targets in synthetic aperture radar (SAR) imagery often yield signatures that are smeared beyond recognition. One mitigation strategy is to attempt to refocus such smears as if the corresponding targets were stationary. This problem is especially difficult for cases in which the target executes time‐varying rotation rates for roll, pitch and...
In this article, we consider two particular bistatic cases which arise in Synthetic Aperture Radar (SAR) imaging: when the transmitter and receiver are moving in the same direction or in the opposite direction and with different speeds. In both cases, we classify the forward operator $\mathcal{F}$ as an FIO with fold/blowdown singularities. Next we...
Citations
... Finally, infrared (IR) cannot be used to measure slick thickness, in general, [8]. Polarimetric imaging, on the other hand, has distinct advantages for a variety of detection and classification problems [32]. This sensor's light reflects directly from the surface, containing the most information on surface oil [17]. ...
Several oil spill disasters have been reported in the last decade, posing a major threat to the marine ecosystem, damaging marine life, vital for protecting the environment, and reducing economic losses. In order to reduce or clean the oil spill, one needs to create a cost-effective oil spill detection system, including its source, the spill extent, the quantity estimate, the range of probable transport paths, and weather and sea conditions. Thermal and polarimetric imagery are emerging sensing modalities that show the potential for enhanced contrast in situations where conventional imaging, such as microwave, hyper-spectral, and visible imaging, has recently been researched. There is a need to compare existing thermal and polarimetric images since there is little work and data in this area. Current studies have shown some improvement in oil spill technique development. Even with the additional availability of new techniques, these steps are limited by cloud cover and lack of contrast. This article will investigate thermal and polarimetric cameras' usage for tracking 3-D oil spills in the sea by developing robust unsupervised oil sensing algorithms. It involves introducing: 1) an oil spill segmentation framework designed for thermal and polarimetric imagery; 2) a multidensity oil spill region enhancement and 3-D thickness visualization algorithm; and 3) a qualitative and quantitative oil spill analysis approach. Comparisons with existing algorithms demonstrate the effectiveness of the proposed algorithms.
... Using SAR images to monitor ships on the sea has the dual value of civilian and military. The civilian field can be used for marine environmental protection, maritime rescue, and fishery supervision; The military field can obtain military intelligence on maritime operations in real time, which is significant for ensuring national security (Lang et al. 2014). ...
As ship target detection technology has high application value in military and civil fields, it is significant to research ship detection in SAR images. Aiming at the complex and diverse backgrounds, significant differences in ship sizes, and real-time detection problems in the ship target detection task of SAR remote sensing images, a lightweight ship detection network based on the YOLOx-Tiny model is proposed. Firstly, a multi-scale ship feature extraction module is proposed, composed of a parallel multi-branch structure connected by a standard convolution layer, asymmetric convolution layer, and dilatation convolution layer with different expansion rates in turn. It makes better use of local features and global features and effectively improves the detection accuracy of multi-scale ship targets; Secondly, to ensure detection performance and eliminate background interference, we propose a whole SAR remote sensing image detection strategy based on an adaptive threshold, which effectively suppresses false alarms caused by background and improves detection speed. The experimental results on two different SAR ship datasets, SSDD and HRSID, show that, compared with several advanced methods, the effectiveness and superiority of the method in this paper are verified, and excellent results are shown in the detection of the whole SAR remote-sensing image. It can provide effective theoretical and technical support for ship detection on platforms with limited computing resources and has good application prospects.
... In recent years, however, many spaceborne SARs have been launched and a number of airborne SARs have also been increased. Correspondingly, several new methods have been proposed for ship classification [197][198][199][200][201][202][203][204]. family, the improved-YOLOv3 model showed the better average precision on both the optical (93.56%) and SAR (95.52%) datasets. ...
... In recent years, however, many spaceborne SARs have been launched and a number of airborne SARs have also been increased. Correspondingly, several new methods have been proposed for ship classification [197][198][199][200][201][202][203][204]. [196]. ...
... The feature-based template matching using the parametric vectors was developed by several institutes, for example, the GMV Spain [197,198], the National University of Defense Technology, China [199], and Lockheed Martin Canada (LMC) [200]. The flowchart on the right of Figure 33 shows the system developed by the GMV Spain. ...
In 1978, the SEASAT satellite was launched, carrying the first civilian synthetic aperture radar (SAR). The mission was the monitoring of ocean: application to land was also studied. Despite its short operational time of 105 days, SEASAT-SAR provided a wealth of information on land and sea, and initiated many spaceborne SAR programs using not only the image intensity data, but also new technologies of interferometric SAR (InSAR) and polarimetric SAR (PolSAR). In recent years, artificial intelligence (AI), such as deep learning, has also attracted much attention. In the present article, a review is given on the imaging processes and analyses of oceanic data using SAR, InSAR, PolSAR data and AI. The selected oceanic phenomena described here include ocean waves, internal waves, oil slicks, currents, bathymetry, ship detection and classification, wind, aquaculture, and sea ice.
... Sugimoto et al. [13] combined Yamaguchi decomposition theory with the CFAR method to accomplish SAR ship detection. Lang et al. [14] proposed a hierarchical SAR ship recognition scheme by extracting texture descriptors to construct a robust ship classification and recognition representation. Leng et al. [15] combined the SAR ship's intensity and location information and proposed a bilateral CFAR method. ...
... Remote Sens. 2022,14, 5148 ...
Synthetic aperture radar (SAR) ship detection has been the focus of many previous studies. Traditional SAR ship detectors face challenges in complex environments due to the limitations of manual feature extraction. With the rise of deep learning (DL) techniques, SAR ship detection based on convolutional neural networks (CNNs) has achieved significant achievements. However, research on CNN-based SAR ship detection has mainly focused on improving detection accuracy, and relatively little research has been conducted on reducing computational complexity. Therefore, this paper proposes a lightweight detector, LssDet, for SAR ship detection. LssDet uses Shufflenet v2, YOLOX PAFPN and YOLOX Decopuled Head as the baseline networks, improving based on the cross sidelobe attention (CSAT) module, the lightweight path aggregation feature pyramid network (L-PAFPN) module and the Focus module. Specifically, the CSAT module is an attention mechanism that enhances the model’s attention to the cross sidelobe region and models the long-range dependence between the channel and spatial information. The L-PAFPN module is a lightweight feature fusion network that achieves excellent performance with little computational effort and a low parametric count. The Focus module is a low-loss feature extraction structure. Experiments showed that on the Sar ship detection dataset(SSDD), LssDet’s computational cost was 2.60 GFlops, the model’s volume was 2.25 M and AP@[0.5:0.95] was 68.1%. On the Large-scale SAR ship detection dataset-v1.0 (LS-SSDD-v1.0), LssDet’s computational cost was 4.49 GFlops, the model’s volume was 2.25 M and AP@[0.5:0.95] was 27.8%. Compared to the baseline network, LssDet had a 3.6% improvement in AP@[0.5:0.95] on the SSDD, and LssDet had a 1.5% improvement in AP@[0.5:0.95] on the LS-SSDD-v1.0. At the same time, LssDet reduced Floating-point operations per second (Flops) by 7.1% and Paraments (Params) by 23.2%. Extensive experiments showed that LssDet achieves excellent detection results with minimal computational complexity. Furthermore, we investigated the effectiveness of the proposed module through ablation experiments.
... In addition, ship detection methods based on ship features are widely utilized. [13], structural features [14], [15], polarization features [16], [17], [18], [19], [20], and scattering features [21], [22], [23]. Recently, Zhu et al. [24] proposed a projection shape template-based ship recognition method. ...
As artificial intelligence continues to advance, deep learning has greatly contributed to the advancement of ship recognition using Synthetic Aperture Radar (SAR) images. Deep learning-based SAR ship recognition performance is largely dependent on the sample set used. SAR ship recognition datasets published in recent years, however, are most derived from a single SAR satellite sensor. It needs to be evaluated and analyzed carefully whether the model trained by a single satellite dataset can still achieve the same accuracy with different SAR satellites. The paper focuses on the following research to address these issues. Firstly, using multiple SAR satellite sensors, we create a new SAR ship dataset (named generalization performance evaluation dataset, GPED) containing multi-resolution and multi-polarization data to examine the generalization performance of the deep learning-based SAR ship recognition method. GPED and a marine target detection dataset (MTDD) are then used to evaluate and analyze the generalization performance of current mainstream deep learning methods. According to the the experiment results, the mean average accuracy (mAP) of the ship recognition model trained on GPED is generally higher than that of MTDD, which proves that GPED has a better generalization performance. Furthermore, SAR ship detection datasets have more samples than ship recognition datasets, which inspired us to use transfer learning to transfer knowledge from ship detection to ship recognition. In this paper, a method for ship recognition based on transfer learning that utilizes the knowledge gained from the ship detection task is proposed. The method includes two modules: pretraining module and fine-tuning module. It can apply samples of unlabeled ship types to ship recognition, thus reducing the number of labeled samples that are required for ship recognition. The experimental results on GPED and MTDD show that our method can achieve good recognition performances.
... ECAUSE of all-weather and day-and-night imaging capability, synthetic aperture radar (SAR) has been used effectively for monitoring the vast ocean where in-situ data collection is difficult [1]. Apart from the natural oceanic phenomena, ship detection is one of the major issues in the maritime domain awareness, and many studies have been reported such as ship detection [2][3][4][5][6][7][8][9] and classification of ships [10][11][12][13]. In addition, multi-look processing was applied to improve ship detection [14,15]. ...
A method for improving the estimation accuracy of the velocity of cruising ships is proposed using synthetic aperture radar (SAR sub-look images in the Spotlight mode. The main purpose of Spotlight SAR is to obtain high resolution utilizing longer integration times than those of other imaging modes, and the proposed method is based on these long integration times. The principal methodology is to produce successive N sub-look images of a cruising ship, where N is more than approximately 10. The positions of the look-1 and look-N sub-images differ by a substantial distance proportional to the cruising speed and the long inter-look time difference. The distance, and hence the velocity of the cruising ship, can be computed from the cross-correlation function of these two sub-look images with improved accuracy compared with other modes. We tested using PALSAR-2 Spotlight sub-images with N= 2, 10 and 20, and the results are compared with the automatic identification system (AIS data. Five images of ships cruising close to the azimuth direction were tested, the best result was obtained for the 10-look images with the average error of 13.8%, followed by 17.9% and 40.5% errors for the 20- and 2-look images respectively. The reason is also given for the best result of the 10-look case over the 20-look case.
... In 2014, Wang et al. [41] also designed a hierarchical ship classifier for COSMO-SkyMed SAR images, where the geometric and backscattering characteristics of various ship types were analyzed and used; however, they merely collected 41 ship samples that would pose a decline in the model robustness. Later, inspired by the above hierarchical classification concept, Lang et al. [42] also designed a hierarchical ship classifier from RadarSAT-2 and TerraSAR-X SAR images, which first extracted dense scale-invariant feature transform (SIFT) [73] as local features and then adopted the k-means clustering algorithm to generate a feature dictionary; finally, a support vector machine (SVM) was used to identify ship categories. ...
Ship classification in synthetic aperture radar (SAR) images is a fundamental and significant step in ocean surveillance. Recently, with the rise of deep learning (DL), modern abstract features from convolutional neural networks (CNNs) have hugely improved SAR ship classification accuracy. However, most existing CNN-based SAR ship classifiers overly rely on abstract features, but uncritically abandon traditional mature hand-crafted features, which may incur some challenges for further improving accuracy. Hence, this article proposes a novel DL network with histogram of oriented gradient (HOG) feature fusion (HOG-ShipCLSNet) for preferable SAR ship classification. In HOG-ShipCLSNet, four mechanisms are proposed to ensure superior classification accuracy, that is, 1) a multiscale classification mechanism (MS-CLS-Mechanism); 2) a global self-attention mechanism (GS-ATT-Mechanism); 3) a fully connected balance mechanism (FC-BAL-Mechanism); and 4) an HOG feature fusion mechanism (HOG-FF-Mechanism). We perform sufficient ablation studies to confirm the effectiveness of these four mechanisms. Finally, our experimental results on two open SAR ship datasets (OpenSARShip and FUSAR-Ship) jointly reveal that HOG-ShipCLSNet dramatically outperforms both modern CNN-based methods and traditional hand-crafted feature methods.
... HR-SAR is a high-resolution SAR ship data set which was collected by Xing et al. [4] from six TerraSAR-X stripmapmode SAR images with 2.0 m × 1.5 m resolution in azimuth and range directions, which has 50 ship samples per class. MR-SAR is a moderate-resolution SAR ship data set which was collected by Lang et al. [5], [8] from eight Radarsat-2 standard-mode VV-polarization images with moderate resolution of about 15 m both in azimuth and range directions, which has 50 ship samples per class. This study assumes that SD and TD have the same classification task, therefore above database contains three same ship classes, i.e., cargos, containers, and oil tankers. ...
... Time complexity: We empirically check the time complexity of all algorithms by running them on the AIS→HR-SAR/MR-SAR tasks. 5 The environment was an Intel Core i7-6800K CPU with 64GB memory. The results in Table III reveal that except its superiority in classification accuracy, GTML also achieved a running time complexity better than other well-performing methods like JGSA, LRSR, LRDR whose MP Table. ...
There are still many challenges to be resolved in the task of ship classification in synthetic aperture radar (SAR) images, such as limited number of labeled samples in SAR domain, large variance in the same subcategory, small variance among different subcategories, etc. Transfer metric learning (TML) has the potential to mitigate those issues in the domain of interest (target domain, TD) by leveraging knowledge/information from other related domains (source domain, SD). In this article, we proposed a novel TML method, termed as geometric transfer metric learning (GTML), which achieves discriminative information preservation (DIP), geometric structure preservation (GSP), and handles the domain shift (DS) simultaneously by integrating pairwise constraints (PC), joint distribution adaptation (JDA), and manifold regularization (MR) into a unified optimization function, aiming to make full use of their complementarity to improve SAR ship classification performance. In practice, we proposed two simple but effective optimization strategies, termed as GTML-A and GTML-R, to construct optimization function. We also proposed two solutions for two typical real-world application scenarios, that is, the task of: 1) zero-labeled sample (ZLS) and 2) scarce-labeled samples (SLS) in SAR domain. The experiments conducted on both tasks show that the proposed GTML outperforms most of state-of-the-art methods. Code is available at
https://github.com/sky-Yongjie-Xu/geometric-transfer-metric-learning
.
... Synthetic Aperture Radar (SAR) can provide day-and-night and weather-independent high resolution images 1 , thus it plays an important role in marine monitoring and maritime traffic supervision 2 . Ship detection in SAR imagery has attracted wide interests and many research works have been carried out 3 . Ship detection in SAR imagery is usually divided into three steps: preprocessing, prescreening and discrimination. ...
Effectively obtaining the location and direction of the ship target is an important prerequisite for maritime traffic management and marine accident rescue. Thanks to the rapid development of the target detection methods based on deep learning, this article proposed a ship target detection method for multiresolution synthetic aperture radar (SAR) images based on improved region convolution neural network (R‐CNN). According to the characteristics of ship target in the SAR images, we make several improvements such as enlarging the input, proposal optimization, database target categorization, and weight balance on the basis of the standard Faster R‐CNN. The experimental results proved that the proposed method can detect target effectively and precisely in complicated scenes of multiresolution SAR images, such as in‐shore and dense targets. It has a good potential in practical application.
... In our previous work [33], the methods were evaluated on the HR-SAR data set. Another data set, MR-SAR is a moderate-resolution SAR ship data set which was collected by Lang et al. [4], [5] from eight Radarsat-2 standard-mode VVpolarization images with moderate resolution of about 15 m both in azimuth and range directions. This database contains three same ship classes as HR-SAR, i.e., cargos, containers, and oil tankers, and fifty ship samples per class. ...
Recent years, the role of the space-borne synthetic aperture radar (SAR) in maritime ship monitoring has become widely accepted. The increase of the number of moderate- to high-resolution SAR images with the advent of the new generation of satellite missions makes it possible to further identify the type of ship (i.e. ship classification) beyond normally provide geographic location of the ship target (i.e. ship detection). However, ship classification in SAR images is by no means a simple task. It remains understudied and still has many challenges to be resolved, such as limited discriminative information provided by SAR images, large variance in the same subcategory and small variance among different subcategories, etc. In this paper, we proposed a novel DML method, termed as distribution shift metric learning (DML-ds), which improves the original Laplacian regularized metric learning (LRML) by adding an inter-class distribution shift (ICDS) regularization term. Extensive experiments and in-depth analysis demonstrate that the proposed DML-ds can effectively increase the inter-class separability and the intra-class compactness, thereby improving the fine-grained ship classification performance in SAR images, and outperforms most of state-of-the-art methods.
