More and more network traffic data have brought great challenge to traditional intrusion detection system. The performance is tightly related selected features classifiers, but feature selection algorithms classification can't perform well in massive environment. Also the raw are imbalanced, which has a serious impact on results. In this paper, we propose novel model utilizing convolutional neural networks (CNNs). We use CNN select from set automatically, cost function weight coefficient of...

The multi-scale object detection, especially small is still a challenging task. This paper proposes an improved detection network based on single shot multibox detector (SSD), and the named as SSD-MSN. SSD-MSN can learn more rich features of objects from enlarged areas, which are clipped raw image. extra contributed to improving performance. includes two subnets: area proposal (APN) network, namely SSD detector. APN used select proposals containing one or areas. predict classification...

With the rapid development of VR technology, capacity audio-video files has increased significantly, demanding higher transmission performance. Traditional file techniques cannot meet these needs due to specificity files. Therefore, research on synchronous over wireless multipath based bidirectional been proposed. Using DNA data storage encoding/decoding, are uniformly encoded and stored. By dividing into two sub-streams considering security needs, a model is established with corresponding...

The research on malware detection enabled by deep learning has become a hot issue in the field of network security. existing methods based suffer from some issues, such as weak ability feature extraction, relatively complex model, and insufficient model generalization. Traditional architectures, convolutional neural networks (CNNs) variants, do not consider spatial hierarchies between features, lose information precise position within region, which is crucial for file specific sections. In...

As a decentralized technology, blockchain has the characteristics of traceability and immutability. Using each model update federation information can be associated with participant, transactions in used to detect fraudulent that attempt tamper data. At same time, is realize decentralization system, which strengthens fault tolerance attack resistance system. Based on this, this chapter designs implements architecture solve problem safe trusted data sharing among participants federated...

As the object detection dataset scale is smaller than image recognition ImageNet scale, transfer learning has become a basic training method for deep models, which pre-trains backbone network of model on an to extract features tasks. However, classification task focuses salient region object, while location edge features, so there certain deviation between extracted by pretrained and those needed localization task. To solve this problem, decoupled self-attention (DSA) module proposed...

RetinaNet proposed Focal Loss for classification task and improved one-stage detectors greatly. However, there is still a gap between it two-stage detectors. We analyze the prediction of find that misalignment localization main factor. Most predicted boxes, whose IoU with ground-truth boxes are greater than 0.5, while their scores lower which shows needs to be optimized. In this paper we an object confidence problem, shares features task. This uses IoUs samples as targets, only losses...

Federated learning (FL) is increasingly challenged by security and privacy concerns, particularly vulnerabilities exposed malicious participants. There remains a gap in effectively countering threats such as model inversion poisoning attacks existing research. To address these challenges, this paper proposes the Effective Private-Protected Learning Aggregation Algorithm (EPFed), framework that utilizes blockchain platform, homomorphic encryption, secret sharing to fortify data computational...

Abstract The traditional rendering technology creates virtual scenes with insufficient fidelity, which are quite different from real scenes. To address this issue, a super-resolution scene based on generalized Huber-MRF image modeling has been studied. This study preprocesses the original through three steps: graying, filtering, and enhancement. is employed for restoration to enhance clarity. Corner features extracted image, Delaunay triangular grid method used construct image's 3D model....

Abstract A steady usage growth of various smart meters/sensors types comes along with the development Smart Grid. Due to that usually produce data continuously, state-of-the-art storage architecture will no longer be sustainable under such a dramatic increase circumstance. It is crucial these shall stored for both short and long term analysis seek valuable information. This paper proposed scalable distributed time series data, further substantiated framework on proof-of-concept testbed using...

With the development of edge computing and cloud in power scenarios, center collects a large amount data from nodes every day, load is overloaded transmission delay increases, making it difficult to store use center. The communication capabilities, storage capabilities face greater challenges. Load most important structural asynchronous heterogeneous electric power. In order reduce generated during process network, compression technology can be used effectively compress data. Before using...

There are many kinds of energy data, how to realize unified storage, processing and sharing data is a big problem. As the national center, State Grid aims build database that can store distributed heterogeneous asynchronous data. The storage image files in will take up lot space system, but not all parts needed. Therefore, it very necessary accurately segment effective area so as achieve purpose compression. This paper proposes Attention U-Net framework, which combines traditional semantic...

Compressing the neural network can significantly reduce its computational complexity, save resources and speed up inference time. However, current compression methods, whether used individually or in combination, often neglect issue of strategy generation, making it challenging to obtain compressed models with smallest accuracy degradation that meet user's deployment requirements. This paper proposes a method for automatically generating strategy, aiming achieve high-performance requirements...

Insulators play a crucial role in both power trans-mission and transformation scenarios, leveraging object detection to analyze insulator defects images holds significant value for production. This paper focuses on the distinct aspects of candidate box processing models, utilizing diffusion model perceptual understanding enhance model's performance defect detection. The proposed demonstrates improvement +6% 2.9% over Faster R-CNN DETR, respectively, transmission Similarly, detection,...

A new joint transfer learning method is designed y integrating the parallel computing methods of MapReduce and Hadoop. At same time, correctness security algorithm are proved theoretically. Build a cluster Hadoop machines, carry out data processing system in case "no decryption", verify results. Experiments show that under framework can greatly reduce time required for encryption, its training accuracy as plaintext.

As the scale of object detection dataset is smaller than that image recognition ImageNet, transfer learning has become a basic training method for deep models, which will pretrain backbone network model on ImageNet to extract features classification and localization subtasks. However, task focuses salient region object, while location edge so there certain deviation between extracted by pretrained used task. In order solve this problem, decoupled self attention(DSA) module proposed one stage...

With the development of IoT technology, smart grid has gradually replaced traditional grid. Smart is convenient and fast. It can provide real-time residential electricity monitoring forecasting, give users better guidance save a lot labor costs. meters send customers’ consumption data to gateway, which aggregates then sends it control center. But in this process, there will be security problem leakage customer’s data. Most current user privacy protection collection schemes use homomorphic...