Recently, significant advancements have been achieved in optimizing algorithms for synthetic aperture radar (SAR) ship detection. Nevertheless, two challenges still impede further research: 1) Mainstream methods, whether anchor-free or anchor-based, adhere to a dense paradigm, leading to substantial redundancy and limited adaptability. 2) Ship targets in SAR images exhibit large shape variations and scale differences, making it difficult to efficiently extract key features from background clutter. To tackle the aforementioned problems, we propose DCEA (DETR with Concentrated dEformable Attention), a query-based method for end-to-end optimization of the current pipeline. First, for the irregular shapes and sparse distribution of ships, the concentrated deformable attention (CDA) is introduced to model the spatial positions of targets, simulating their geometric transformations with precision. Second, an attention-wise propagation module (APM) is designed to integrate local fine-grained information with global semantic information, improving the detection performance for objects across diverse scales. Finally, due to the lack of information exchange between object queries, a dimension-wise information mixing (DIM) module is employed to incorporate key information from various dimensions to enhance their representation capability. To validate the superior performance of DCEA, we conduct extensive experiments on multiple public datasets, achieving mean average precision (mAP) scores of 0.991, 0.929, and 0.962 on the SSDD, HRSID, and SAR-Ship-Dataset, respectively, with a model size of only 14.34M parameters and 44.4 GFLOPs.

Capsule Network (CapsNet) is a novel structure for deep neural network, mapping the region of target instance to vectors and matrices rather than scalars. This process is enabled by dynamic routing algorithm which help CapsNet to achieve more robust capacity with fewer parameters than traditional CNNs. However, one drawback of Capsule is that it turns to account for everything in the image, which leads to a poor performance when the backgrounds are too varied to model with a reasonable sized net. We proposed a multi-lane capsule network with strict-squash (MLSCN) to solve this problem. In MLSCN, we designed a novel Capsule based network structure, replaced the Squash function and optimized the implementation of dropout. To validate modification proposed in the paper, we conducted extensive experiments on three public datasets which include MNIST, affNIST and CIFAR10. Ablation experiments were also conducted to analyze contributions of each component of MLSCN. Experimental results show that MLSCN outperforms the original CapsNet in multiple benchmarks. Our model boosts the classification accuracy of CIFAR10 about 17% with negligible parameter increase compare to original CapsNet. Besides, the proposed model achieves an accuracy of 65.37%, while original CapsNet is only 41.62% on MNIST-BC.

In the era of mobile big data, data driven intelligent Internet of Things (IoT) applications are becoming widespread, and knowledge-based reasoning is one of the essential tasks of these applications. While most knowledge-based reasoning work is conducted with knowledge graph, ontology-based reasoning method can inherently achieve higher level intelligence by leveraging both explicit and tacit knowledge in specific domains, and its performance is determined by precise refinement of the inference rules. However, most ontology-based reasoning work concentrates on semantic reasoning in a single ontology, and fail to utilize association of multiple ontologies in various domains to extend reasoning capacity. This is even the case for the IoT applications where knowledge from multiple domains needs to be utilized. To overcome this issue, we propose a deep learning-based method to associate multiple ontology rule bases, thereby discover new inference rules. In our method, we first use a regression tree model to determine the threshold value for parameters in inference rules that constitute the ontology rule base, avoiding the influence of uncertainty factors on knowledge reasoning results. Then, a two-way GRU (Gated Recurrent Unit) neural network with attention mechanism is used to discover semantic relations among the rule bases of ontologies. Therefore, the association of multiple ontology rule bases is realized, and the rule base of knowledge reasoning is expanded by acquiring some unspecified rules. To the best our knowledge, this work is the first one to leverage deep learning in reasoning with multiple ontologies. In order to verify the effectiveness of our method, we apply it in a real traffic safety monitoring application by relating rule bases of a vehicle ontology and a traffic management ontology, and achieve effective knowledge reasoning.