Pixel-level image fusion is an effective way to fully exploit the rich texture information of visible images and the salient target characteristics of infrared images. With the development of deep learning technology in recent years, the image fusion algorithm based on this method has also achieved great success. However, owing to the lack of sufficient and reliable paired data and a nonexistent ideal fusion result as supervision, it is difficult to design a precise network training mode. Moreover, the manual fusion strategy has difficulty ensuring the full use of information, which easily causes redundancy and omittance. To solve the above problems, this paper proposes a multi-stage visible and infrared image fusion network based on an attention mechanism (MSFAM). Our method stabilizes the training process through multi-stage training and enhances features by the learning attention fusion block. To improve the network effect, we further design a Semantic Constraint module and Push–Pull loss function for the fusion task. Compared with several recently used methods, the qualitative comparison intuitively shows more beautiful and natural fusion results by our model with a stronger applicability. For quantitative experiments, MSFAM achieves the best results in three of the six frequently used metrics in fusion tasks, while other methods only obtain good scores on a single metric or a few metrics. Besides, a commonly used high-level semantic task, i.e., object detection, is used to prove its greater benefits for downstream tasks compared with singlelight images and fusion results by existing methods. All these experiments prove the superiority and effectiveness of our algorithm.

Name disambiguation has long been a significant issue in many fields, such as literature management and social analysis. In recent years, methods based on graph networks have performed well in name disambiguation, but these works have rarely used heterogeneous graphs to capture relationships between nodes. Heterogeneous graphs can extract more comprehensive relationship information so that more accurate node embedding can be learned. Therefore, a Dual-Channel Heterogeneous Graph Network is proposed to solve the name disambiguation problem. We use the heterogeneous graph network to capture various node information to ensure that our method can learn more accurate data structure information. In addition, we use fastText to extract the semantic information of the data. Then, a clustering method based on DBSCAN is used to classify academic papers by different authors into different clusters. In many experiments based on real datasets, our method achieved high accuracy, which proves its effectiveness.

Infrared (IR) and visible (VI) image fusion play an important role in improving the sea scene perception and ship target detection. Although there have been many studies on image fusion considering the sea scene characteristics, there has still been no adequate extraction method for detailed information of a ship target in a sea scene. To overcome this shortcoming, this paper proposes a fusion method based on a scale‐aware feature decomposition, which can accurately extract features of a ship target at different scales. First, a hybrid feature decomposition method based on the scale‐aware structure‐preserving filter and Gaussian filter is designed. The proposed method separated source images into region, structure, and texture layers, and thus achieved a finer‐scale division than traditional multiscale decomposition methods in the sea‐clutter background. Then, further decomposition of the texture layer was performed by the shearlet transform to obtain the directional texture feature. According to the characteristics of each layer, different strategies were adopted for the fusion, and the weighting factor was used to obtain the final fusion result. Experimental results indicated that the proposed method could achieve better subjective and objective results than current state‐of‐the‐art methods. This method introduces a scale‐aware edge preserving filter to improve the ability of detail extraction and is suitable for image fusion of ship targets.

With the development of technology, image and other data are developing in the direction of higher resolution and higher frame rate, which puts forward higher requirements for the speed of big data real-time storage systems. At the same time, in order to facilitate the use of outdoor and other practical scenes, more miniaturization and integration of storage system are required.In order to solve the problems of high-speed real-time storage, device integration and miniaturization, this paper designs an NVME SSD real-time high-speed storage system based on ZYNQMP SoC. The ZYNQMP SoC is connected to the NVME SSD through PCIE interface, and the NVME protocol is migrated to complete the interaction with the NVME SSD, so as to achieve higher rate of read and write. Migrate the exFAT file system to manage files in NVME SSD.The results show that the system can read and write NVME SSD at a speed of 1600MB/s. After NVME SSD is inserted into a PC, exFAT file system can be recognized and files can be read and written normally, and real-time, high-speed and stable storage is realized.

Aiming at the problems of low efficiency of feature extraction and insufficient use of feature correlation in image retrieval based on deep hashing, an image retrieval method Dual Learning Hashing(DLH) that combines feature layer learning and hash layer learning is proposed. In view of the problem that existing methods usually extract local features with intensive attention mechanism by focusing on dense local regions, these local regions cannot contain different local information, a module is developed to learn differentiated local features by locating the peaks of non-overlapping subdomains in the feature map. In order to make full use of semantic information and generate high quality hash code, quantization function and probabilistic semantic-preserving function are designed in the hash layer. Finally, combined with the two parts of learning, the total loss function is given. A large number of comparative experiments have been carried out on three widely used datasets. The experimental results show that compared with other advanced deep hash methods, DLH has achieved better retrieval performance.

Ship target detection on the sea surface is one of the most common scenes in the field of target detection. The accuracy of target detection plays an important role in the field of sea rescue and defense. However, complex infrared sea surface scenes, such as island shore, fish scale wave, sea surface bright band and other disturbances, bring great challenges to target detection. In this paper, we propose an improved YOLOV5 model by analyzing the characteristics of infrared image imaging and ship target on the sea surface. Aiming at the problem of information loss of small targets in the deep layer of the network, we redesigned the backbone network, which was composed of four Multi-scale residual blocks, and each block was connected by CBAM (Convolutional Block Attention Module) attention mechanism to improve information transmission and fusion between different feature layers. Aiming at the problem of complex sea surface scene interference, we use FPN and PAN (Feature Pyramid Network and Personal Area Network) mechanism to construct feature fusion network in neck, and add shallow feature fusion in FPN. Our model achieves good performance both in terms of latency and accuracy. In a self-collected sea surface scene dataset, multiple SOTAs for detection tasks are compared and the results demonstrate the superiority of the proposed method.