Search

Multioriented object detection, an important yet challenging task because of the bird's-eye-view perspective, complex background, and densely packed objects, is in the spotlight of detection in remote sensing images. Although existing methods have recently experienced substantial progress based on oriented head, they learn little about essential rotation invariance of the objects. In this article, a novel framework is proposed that can learn high-quality rotation invariance features of the multioriented objects by three measures. Given a remote sensing image, the multiscale semantic segmentation feature fusion module first merges the global semantic segmentation features predicted by the semantic segmentation branch and the multiscale features extracted by the backbone with FPN in order to distinguish complex background. Then, the discriminative features are used by rotation mainstream, whose structure is similar to cascade R-CNN and can extract higher quality rotation invariance features and predict more accurate location information by adaptively adjusting the distribution of the samples through progressive intersection over union thresholds. And in order to improve the performance of mainstream to predict more accurate oriented bounding box, the horizontal tributaries that can fully leverage the reciprocal relationship between the oriented detection and horizontal detection were added to the latter two stages. Extensive experiments on three public datasets for remote sensing images, i.e., Gaofen Airplane, HRSC2016, and DOTA demonstrate that without bells and whistles, the proposed method achieves superior performances compared with the existing state-of-the-art methods for multioriented detection. Moreover, our overall system achieves 59.264% mAP of airplane Detection in 2020 Gaofen challenge, ranking third in the final.

Most deep-learning-based target detection methods have high computational complexity and memory consumption, and they are difficult to deploy on edge devices with limited computing resources and memory. To tackle this problem, this article proposes to learn a lightweight detector named Light-YOLOv4, and it is obtained from YOLOv4 through model compression. To this end, first, we perform sparsity training by applying L1 regularization to the channel scaling factors, so the less important channels and layers can be found. Second, channel pruning and layer pruning are enforced on the network to prune the less important parts, which could significantly reduce network's width and depth. Third, the pruned model is retrained with a knowledge distillation method to improve the detection accuracy. Fourth, the model is quantized from FP32 to FP16, and it could further accelerate the model with almost no loss of detection accuracy. Finally, to evaluate Light-YOLOv4's performance on edge devices, Light-YOLOv4 is deployed on NVIDIA Jetson TX2. Experiments on the SAR ship detection dataset (SSDD) demonstrate that the model size, parameter size, and FLOPs of Light-YOLOv4 have been reduced by 98.63%, 98.66%, and 91.30% compared with YOLOv4, and the detection speed has been increased to 4.2×. While the detection accuracy of Light-YOLOv4 is only slightly reduced, for example, the mAP has only reduced by 0.013. Besides, experiments on the Gaofen Airplane dataset also prove the feasibility of Light-YOLOv4. Moreover, compared with other state-of-the-art methods, such as SSD and FPN, Light-YOLOv4 is more suitable for edge devices.

Ship detection and recognition in remote sensing images have important significance in military and civilian applications. Traditional methods have insufficient generalization ability in complicated scenes. The Faster R-CNN-based methods cannot predict the orientation of the ship. The R2CNN-based methods can predict the orientation of the ship but not considerate the scale of object in classification. In order to solve the problems mentioned above, this article proposes a scale enhancement rotating Cascade R-CNN network (SER-Cascade). Using the multistage network of rotating Cascade R-CNN, the output of the previous stage is fed to current stage, which can effectively regress the orientation of the ship. To improve the recognition performance of multi-class ships, a novel RoI pooling method is proposed in this article, in which the scale information is enhanced and context information is reserved. To evaluate the proposed networks, a dataset named HR-SHIP-15 that currently contains 15 categories of ship targets has been produced for ship recognition. Experiments are conducted on HR-SHIP-15 dataset, and the results verify that the proposed method has state-of-the-art performance.

Water segmentation is a fundamental step for the information processing of SAR image, which plays an important role in ship detection, disaster monitoring and other applications. Because of the complexity of scenario in the SAR images, water segmentation of SAR image is a challenging task. Fewer convolutional neural networks (CNNs) have been developed for SAR image water segmentation in recent years, the accuracy and speed of CNN for water segmentation can be further improved. In this paper, we established a SAR water segmentation dataset based on the GF3 satellite data. then an improved water segmentation network based on Bilateral Segmentation Network (BiSeNet) is proposed. Further we propose a loss function based on edge area and a novel training data generation method to improve the segmentation ability of the network. Experimental results based on water segmentation dataset show that the proposed segmentation method has better segmentation accuracy and speed.

Recently, complex signal kurtosis (CSK) is found to be a vital indicator of ship detection in synthetic aperture radar (SAR) images. Since CSK can be used as an indicator of ship detection, we consider it can also be used to indicate ship target signature. One of the basic rationales is that the higher the CSK is, the easier it is to be detected and thus the more obvious its characteristics are as a ship. Presence of sidelobes and defocusing are two common problems that affect the quality of ship targets. They are discussed in this paper from the perspective of CSK. Experimental results show that CSK can be used to indicate and improve ship quality. We believe that ship detection and recognition can benefit from the CSK indicator.

Convolution neural networks doing well in the field of image classification obeys the precondition that test images are similar to the train images. In SAR vehicle image classification field, the recognition rate drops rapidly as the depression angle increasing. The difference of SAR imaging in depression angle obviously affects the classification accuracy of SAR images. In order to overcome the low accuracy in the classification of three categories of SAR vehicles, experiments with a train set composed of depression angle 17° images and a test set composed of depression angle 45° from MSTAR database are explored. To make the test and train set become more similar, the test images are stretched in both horizontal and vertical directions. The stretched method is verified through the cross-correlation coefficient experiments. Meanwhile, an improved structure of convolutional neural network is adopted in the paper to act as the classifier. The recognition accuracy is compared with other traditional machine learning methods and the A-ConvNets method. The results show that the stretching method and the proposed CNN exceed most of existing classifiers in SAR vehicle recognition.

The overestimation of volume scattering (OVS) generally leads to misinterpretations of scattering mechanism since it widely exists in model-based polarimetric synthetic aperture radar (PoISAR) target decomposition. Based on the general four-component scattering power decomposition (G4U), this paper proposes an extended G4U employing two angle compensations (the orientation and helix angle compensations) and two refined models (the cross scattering model and the GVSM). Accordingly, two modified branch conditions are developed to triage land covers and predetermine scattering mechanism, respectively. The proposed method completely utilizes the polarimetric information and relaxes the OVS remarkably. Two different real L-band airborne PolSAR images acquired over different test sites are used to quantitatively and qualitatively demonstrate the effectiveness of the proposed method.

For optimizing the linear array SAR (LASAR) system, the non-uniform linear arrays are usually used to reduce the number of array elements required. However, the data acquisition may be irregular sampling, which is impractical for traditional imaging algorithms. In this paper, based on compressed sensing (CS) approach for the non-uniform linear arrays, we proposed an imaging algorithm for the irregular sampling data. The contributions are twofold. Firstly, it employs the redundant frame to provide more samples and achieve a satisfactory resolution. Secondly, in order to decrease the massive calculations, it combines the truncated singular value decomposition (TSVD) and CS algorithm, while the imaging quality would not be contaminated. Both the theory analysis and simulation results are presented to demonstrate the validity of the proposed algorithm.

The mechanical dynamics in addition to its bulk translation of the target or any structure on the target is called micro-motion, which yields new features in the target's signature that are distinct from its signature in the absence of micro-motion. Micro-motion evokes a frequency modulation in radar echo known as micro-Doppler (m-D) effect which may help to detect specific intrinsic structures of the target, leading a potential method to perform target discrimination and identification by extracting micro-Doppler features. At present, it has drawn a lot of attention to extract the micro-motion target's m-D information for target classification and identification. The various m-D classification approaches require first the extraction of salient features from the radar signal. The micro-Doppler usually manifests curves characteristic in the time-frequency (T-F) domain. Thus the features are calculated from the joint T-F domain. Most of these techniques treat the spectrogram in T-F domain as an image, and obtain features through some image processing techniques. In this paper, we investigated statistical classification and recognition methods for target classification using their micro-Doppler signatures. In our work, micro-Doppler signatures for targets represented by point scattering model with four different micro-motions (Vibration, Coning, Spinning, and Precession) are studied. We propose use of principle component analysis (PCA) and 2-D PCA as the data driven feature extraction approaches that captures vital statistics of the input at a reduced dimension. Simulation analysis by using the simulated data is performed to confirm the effectiveness of the proposal. Experiment results show that with the proposed methods, perfect classification of four different motions can be attained when training and testing set has data from different targets.

Micro-Doppler effect refers to the additional Doppler frequencies around main Doppler frequency which is caused by the micro-motion. Radar targets' motion parameters are useful for characterizing the target and become inputs to automatic target recognition algorithms, which means parameters estimation is fundamental and vital for the target recognition. To improve the instance frequency (IF) estimation accuracy, this paper proposes a new time-frequency analysis (TFA) method combining Choi-Williams Distribution (CWD) and cross Wigner-Ville distribution (XWVD). Firstly, Hadamard product of T-F distribution matrix from T-F distributions of CWD and XWVD is calculated. Then we apply peak value detection (PVD) to get the IF estimation. The simulation results prove its effectiveness and good performance under low SNR. The proposed method not only has better T-F congregation, but also works well even at low SNR, compared to the traditional method of WVD based PVD.

Target segmentation is the basic work of the synthetic aperture radar (SAR) image interpretation. As for aircraft, it is difficult to accurately extract the target from the background due to its complex structure. This paper proposes a refined segmentation method for aircraft target, in which the ratio of exponentially weighted averages (ROEWA) operator is used to produce the edge map and the shape prior is utilized by the point distribution model (PDM). The Catmull-Rom (CR) spline is used to build shape samples. Combining the shape prior and the gradient vector flow (GVF) snake model, the refined segmentation of the aircraft is achieved. The experimental results based on Ku-band airborne SAR data illustrate the effective performance of the proposed method on refined segmentation for aircraft targets.

The mechanical dynamics in addition to its bulk translation of the target or any structure on the target is called micro-motion. Micro-motion evokes a frequency modulation in radar echo known as micro-Doppler (m-D) effect which may help to detect specific structures of a target with micro-motion, leading a potential way to realize target identification by extracting the micro-Doppler features. In most literatures, the micro-motions were modeled as the sinusoidal forms. In fact the micro-motions in reality should include more types of movement forms besides the sinusoidal one. This paper derives the foundation of universal micro-motion model for radar target other than the popular sinusoidal model. The mathematical formulas of micro-Doppler induced by targets with arbitrary micro-motion are derived, and some examples of several basic and typical micro-motion types (vibration, rotation, sinusoidal, pendulum) are given. At last the simulation studies based on the point scattering model are conducted to validate the proposed theoretical model.

Since using conventional constant false alarm ratio (CFAR) algorithms for weak target detection in synthetic aperture radar (SAR) images cause lots of false alarms, this paper proposes a novel detection method based on the Itti visual saliency model. The authors first obtain the feature maps from a SAR image including local variance, frequency of intensity values and global contrast. After Gaussian blurring and normalizing, feature maps are merged using a nonlinear strategy at the original image scale to form the saliency map. As detection results, saliency areas are extracted from the saliency map by combining the focus of attention (FOA) metastasis and region growth methods. Compared with other three methods through experiments, the proposed algorithm performs the best both in accuracy and time complexity.

The micro-Doppler induced by micro-motion in radar target represents an important characteristic made available for target recognition, which is usually manifest a sinusoid in the time-frequency (T-F) domain. Inverse Radon transform has been validated as an effective sinusoid detection method. Considered on the condition that multiple sinusoids existed in the T-F plane especially one is weaker relatively, this paper proposes an enhancing normalized inverse Radon transform (ENIRT) method to detect sinusoids in T-F plane as to extract the micro-Doppler parameters reflecting the target's intrinsic character. The improved method enhances weak sinusoidal curves firstly, and then inverse Radon transformed data is normalized by a normalization matrix. Preliminary results illustrates that the proposed algorithm improves sinusoidal curves detection performance in T-F plane.

It is the basis of realization ship recognition to accurately extract geometric parameters of the ship target in synthetic aperture radar (SAR) images. Due to the unique SAR imaging mechanism, speckle noise, azimuth ambiguity and side lobe effect seriously impact on geometric parameter estimation of the ship target. Therefore, a method is presented to extract geometric parameters of the ship based on ellipse fitting and the gradient vector flow (GVF) snake with shape priors.

This paper proposes a novel registration method for optical and SAR images which is based on straight line features and mutual information. Firstly, different edge detectors are employed to detect the line segments in both optical and SAR images respectively. Then, through the Hough transform and a straight line fitting and filtration method, the main straight lines of each image are extracted and their intersections are obtained and taken as the candidate matching points. With the RANSAC (RANdom SAmpling Consensus) method, corresponding point pairs (CPPs) are found with these candidate points and a coarse registration between the heterogeneous images is implemented. At last, by using the mutual information of the separated patches generated from the coarse registered images, a fine registration result is finally achieved. The experiment with a pair of X-band air-borne SAR and optical images validates the efficiency and precision of the proposed method.

Most of methods for change detection of floods in SAR images are based on the difference image (DI), which is from comparing the flood image and the reference image straightforwardly. DI, however, is often distorted by the interferences of the complicated reference image. Meanwhile, the spatial contextual information of floods is difficult to be effectively used. In response to these problems, a novel change detection method by removing the interference of the reference image is proposed. Firstly, the difference image is segmented to derive the initial change mask, i.e. a part of flood areas. Secondly, the whole flood areas are gained by region growing directly in the flood image. Experimental results on the real ERS SAR dataset validate its effectiveness on both the quantitative and subjective aspects.

In this paper, a new azimuth ambiguities removal method is introduced for ship detection by Time-Frequency (TF) analysis. A TF coherence indicator is proposed to filter ghost echoes due to the different TF coherence characteristics between real ship target echoes and ambiguous ones. The effectiveness of this proposed TF coherence indicator for ship detection is demonstrated using single polarimetric spaceborne TerraSAR-X coherent data over the test sea/ocean site in Hongkong, China.

This paper presents a method in finding tie points automatically in synthetic aperture radar (SAR) image pairs based on the extracted closed regions. There are mainly three steps during this process. The first step is to extract the closed regions in the SAR image with an image segmentation approach deducted by the geodesic active contour (GAC) model. Then, a polygonal approximation process is adopted to locate the feature points on the boundaries of these regions. With the obtained feature points, geometric hashing theory is employed to match these feature points as the tie points. A pair of simulated SAR images and a pair of high-resolution airborne SAR images are used to test and evaluate the proposed method. The experimental results show that the proposed method is effective and appropriate for the acquisitions of tie points in SAR image pairs.

The paper presents a method to detect Targets in synthetic aperture radar (SAR) images at a specified scene. Using the corresponding optical image of the specified scene, the potential regions of interests for Targets can be extracted off-line as a scene mask. Then, the SAR image can be co-registered with this scene mask and a change detector is used to detect the Targets only in the potential regions of the SAR image. Finally, the Targets can be located and detected with the change detection results. The experimental result shows that the proposed method can detect and locate airplanes well at the airport with a high resolution SAR image, and it obtains a high positive detection ratio and evidently decreases the probability of the false alarm. As a result, the proposed method fusing the SAR and optical data is appropriate in detecting airplanes at the airports, vehicles on the roads, ships in the harbours, and so on.