Infrared and visible dual-modality vision tasks such as semantic segmentation, object detection, and salient object detection can achieve robust performance even in extreme scenes by leveraging complementary information. However, most existing image fusion-based methods and task-specific frameworks exhibit limited generalization across multiple tasks. Moreover, summing the general representations obtained from foundation models poses challenges, including insufficient semantic information mining and feature fusion. In this paper, we propose a fusion-enhanced network, which effectively enriches semantic information and integrates features based on the complementary characteristics of infrared and visible modalities. The proposed network can extend to high-level vision tasks, showing strong generalization capabilities. Firstly, we adopt the infrared and visible foundation models to extract the general representations. Then, to enrich the semantic information of these general representations for high-level vision tasks, we design the feature enhancement module and the token enhancement module for feature maps and tokens, respectively. Besides, the attention-guided fusion module is proposed for effective fusion by exploring the complementary information of two modalities. Moreover, we adopt the cutout&mix augmentation strategy to conduct the data augmentation, which further improves the ability of the model to mine the regional complementarity between the two modalities. Extensive experiments show that the proposed method outperforms state-of-the-art dual-modality methods in the semantic segmentation, object detection, and salient object detection tasks.

To meet novel services and networking requirements towards the next generation applications, intent-driven network is proposed as a promising networking paradigm. It is with capabilities of intent refinement, policy generation, and state awareness. And these distinctive capabilities contribute to its wide applications to the next generation networks. However, current researches lack a generalization model of intent refinement. Additionally, it is difficult to extract available knowledge from huge raw data of the network status, and guarantee the precise generation of network policies. To solve these challenges, we present a knowledge graph-enabled intent-driven network with the digital twin, which is termed as KID in this work. In the KID, knowledge graph is utilized to represent user intents, abstract network status, and express network policies. And the digital twin is applied to validate intents as well as abstract the physical network. The KID enhances the capabilities of intent-driven networks to refine intents, contributing to the continuous assurance of accurate intent fulfillment. Finally, we present a proof of concept implementation of the KID. Simulation results verify the feasibility and effectiveness of the presented KID framework.

Speaker verification has been widely used in many authentication scenarios.However, training models for speaker verification requires large amounts of data and computing power, so users often use untrustworthy third-party data or deploy thirdparty models directly, which may create security risks.In this paper, we propose a backdoor attack for the above scenario.Specifically, for the Siamese network in the speaker verification system, we try to implant a universal identity in the model that can simulate any enrolled speaker and pass the verification.So the attacker does not need to know the victim, which makes the attack more flexible and stealthy.In addition, we design and compare three ways of selecting attacker utterances and two ways of poisoned training for the GE2E loss function in different scenarios.The results on the TIMIT and Voxceleb1 datasets show that our approach can achieve a high attack success rate while guaranteeing the normal verification accuracy.Our work reveals the vulnerability of the speaker verification system and provides a new perspective to further improve the robustness of the system.

Collaborative filtering is a popular tool for recommendation systems. However, collaborative filtering technologies often suffer from high time complexity, the cold-start problem, and low coverage. Recent research shows that social networks and trust-aware methods can effectively solve these proble ms. Therefore, we propose a Trust Domain Expert Collaborative Filtering recommendation system. First, we divide the user item rating matrix into multiple sub-matrices based on the domain attributes of each item. For each sub-matrix, we then use domain experts to construct a user–expert trust matrix. Finally, combined with the target user’s domain of interest, we predict their missing ratings. Experimental results show that this method not only improves the accuracy and recommended coverage of collaborative filtering-based methods, but also reduces the computation time.

In our study, multi-temporal HJ-CCD data are used to identify and monitor the powdery mildew (Blumeria graminis f. sp. triti) on winter wheat on a regional scale. A total of five sensitive vegetation indices (NDVI, RDVI, SAVI, TVI, and MSAVI) are specifically derived to construct the minoring models, which can reflect the leaf area index (LAI), chlorophyll (Chl) and changes of crop canopy structure. Nine single-variable monitoring models based on single-period image and two multi-variable monitoring models based on three-period images are respectively constructed to investigate the occurrence of the disease. The results show that the overall accuracies (OA) are located during 71.9%-78.9% for the nine single-variable models in which that of MSAVI is the highest. In the four original bands, that of RNIR is the highest. Conversely, the accuracy of MSF model based on multi-temporal HJ-CCD imagery has been significantly improved. To improve the monitoring accuracy, AdaBoost model is used to optimize the classification model of MSF. It is obvious that the OA of MSF-AdaBoost model is the higher, and the OE and CE are respectively 9.5% and 7.3%, which are lower than the other models.

It is highly necessary to merge high spatial resolution panchromatic images with high spectral resolution multispectral images in image processing tasks and thematic applications. WorldView-3 (WV-3) imagery was investigated for generating pan-sharped multispectral imagery. Brovey and NNDiffuse pan sharpening algorithms were comparatively used to perform the image fusion and the quality was also assessed. The results show that, in comparison with Brovey, NNDiffuse pan sharpened image can generally maintain spectral and texture information. Specifically, the spectral consistency was evaluated using four typical land cover types with a performance of vegetation > waterbody > bare land > built-up area in ascending order. The relative errors were 5.13%, 5.26%, 5.99%, 28.28%, respectively. Entropy was used to evaluate the texture feature and NNDiffuse fused image had extremely similar values in Max, Mean and Stdev compared to original panchromatic image, while there were greater differences for Brovey fused image in Mean and Stdev and the relative errors were -16.04% and 64.88%. Additionally, WV-3 has more clear spatial stereo and edges compared to GF-1 with the same spatial resolution.

Executing ad hoc queries against large databases can be prohibitively expensive. Exploratory analysis of data may not require exact answers to queries, however: results based on sampling the data are often satisfactory. Supporting sampling as a primitive SQL operator turns out to be difficult because sampling does not commute with many SQL operators.In this paper, we describe an implementation in IBM® DB2® Universal Database (UDB) of a sampling operator that commutes with some SQL operators. As a result, the query with the sampling operator always returns a random sample of the answers and in many cases runs faster than it would have without such an operator.