No abstract

Multi-scale object detection within Synthetic Aperture Radar (SAR) images has become a research hotspot in SAR image interpretation. Over the past few years, CNN-based detectors have advanced sharply in SAR object detection. However, the state-of-the-art detection methods are continuously limited in Feature Pyramid Network (FPN) designing and detection anchor setting aspects due to feature misalignment and targets’ appearance variation (i.e., scale change, aspect ratio change). To address the mentioned limitations, a scale-aware feature pyramid network (SARFNet) is proposed in this study, which comprises a scale-adaptive feature extraction module and a learnable anchor assignment strategy. To be specific, an enhanced feature pyramid sub-network is developed by introducing a feature alignment module to estimate the pixel offset and contextually align the high-level features. Moreover, a scale-equalizing pyramid convolution is built through 3-D convolution within the feature pyramid to improve inter-scale correlation at different feature levels. Furthermore, a self-learning anchor assignment is set to update hand-crafted anchor assignments to learnable anchor/feature configuration. By using the dynamic anchors, the detector of this study is capable of flexibly matching the target with different appearance changes. According to extensive experiments on public SAR image data sets (SSDD and HRSID), our algorithm is demonstrated to outperform existing boat detectors.

No abstract

The existing sparse representation-based visual trackers mostly suffer from both being time consuming and having poor robustness problems. To address these issues, a novel tracking method is presented via combining sparse representation and an emerging learning technique, namely extreme learning machine (ELM). Specifically, visual tracking can be divided into two consecutive processes. Firstly, ELM is utilized to find the optimal separate hyperplane between the target observations and background ones. Thus, the trained ELM classification function is able to remove most of the candidate samples related to background contents efficiently, thereby reducing the total computational cost of the following sparse representation. Secondly, to further combine ELM and sparse representation, the resultant confidence values (i.e., probabilities to be a target) of samples on the ELM classification function are used to construct a new manifold learning constraint term of the sparse representation framework, which tends to achieve robuster results. Moreover, the accelerated proximal gradient method is used for deriving the optimal solution (in matrix form) of the constrained sparse tracking model. Additionally, the matrix form solution allows the candidate samples to be calculated in parallel, thereby leading to a higher efficiency. Experiments demonstrate the effectiveness of the proposed tracker.