In the realm of autonomous driving, LiDAR and camera sensors play an indispensable role, furnishing pivotal observational data for critical task precise 3D object detection. Existing fusion algorithms effectively utilize complementary from both sensors. However, these methods typically concatenate raw point cloud pixel-level image features, unfortunately, a process that introduces errors results in loss information embedded each modality. To mitigate problem lost feature information, this...

At present, the effect of object detection algorithm in small is very poor, mainly because low-level network lacks semantic information and characteristic expressed by inspection data lack. In view above difficulties, this paper proposes a based on multi-scale feature fusion. By learning shallow features at level deep level, proposed scheme focuses fusion concrete abstract features. It constructs detector (MSFYOLO) considers relationship between single local environment. Combining global...

Unmanned aerial vehicle (UAV) image object detection, in recent years, has been receiving increasing attention for its wide application military and civil fields. Current detection methods perform well generic scenarios, while vast small objects extremely dense distribution UAV images make it difficult to capture them, resulting sub-optimal performance. In this paper, we propose a framework APNet, which addresses the issue mentioned above by fine-grain deformable convolution (DC) effective...

In video surveillance, there are many interference factors such as target changes, complex scenes, and deformation in the moving object tracking. order to resolve this issue, based on comparative analysis of several common detection methods, a recognition algorithm combined frame difference with background subtraction is presented paper. algorithm, we first calculate average values gray continuous multi-frame image dynamic image, then get obtained by statistical sequence, that is,...

When the object detection algorithm is applied to bird protection project, there are many problems like large model parameters, high similarity between species and single sample scene. In order further improve accuracy stability of model, a multi-object for fine-grained birds proposed. Firstly, introduces Depthwise separable convolution into feature extraction layer YOLOv3 algorithm. The process divided two parts: deep point-by-point convolution. separation intra-channel inter-channel...

When Vertical Federated Learning is used to classify tasks, a large number of invalid parameters are produced. In view the above problems, we propose general method parameter sharing and gradient compression for both sides communication, improve homomorphic encryption transfer parameters. The experimental results show that evaluation index classification model greatly improved compared with traditional longitudinal federated learning logic regression algorithm.

A short-term load forecasting model is adopted with a combined method. The not only summarizes virtues and defects of neural networks fuzzy system, but also considers that power system has characteristics basic heft variability heft. It uses learned capability to complete work for load. Other effect factors cause variety are unconsidered in networks. For affected by many factors, such as weather, data types holidays, membership functions rules base constructed logic which used correct method...

For implementing the data safe transmission in Internet, a mechanism which base on RSA and triple DES algorithm is put forward. The makes full use of advantage RSA. Because encryption speed faster than for long plaintext, distribute key safely easily. Date security far higher algorithm. Digital abstract MD5 adopted this mechanism. Through comparing digital signature transmitted by dispatcher result plaintext got receiver through algorithm, can be guaranteed. This realizes confidentiality,...

When the support vector machine is used for load forecasting, input samples of have important effect to forecasting results. Support can study any non-linear relation, but if a group non-distinct variables are selected as variable set, training time lengthened and errors become bigger. The relation be effectively explained only when appropriate found. In this paper, correlation coefficient idea selection short-term model. values, which bigger with expectation output chosen from factor sets...

For enhancing fault diagnosis precision, the wavelet packet analysis and least squares support vector machine are combined effectively. First, signals decomposed in arbitrary minute frequency bands by use of technique. Doing energy calculation these to form eigenvectors is more reasonable. And then a model presented. When used diagnosis, Fibonacci symmetry searching algorithm simplified improved. It presented choose parameter kernel function on dynamic, which enhances preciseness rate...

In some developed countries, the automatic vehicle recognition is a quite mature technology. This paper applies multi-classification method based on support vector machine (SVM) to recognition. Support machine, appeared recently, new theory and technology in filed of pattern has shown excellent performance practice. was proposed basing structural risk minimization (SRM) place experiential (ERM), thus it good generalization capability. By mapping input data into high dimensional...

This paper proposes a small object detection algorithm based on multi-scale feature fusion. By learning shallow features at the level and deep level, proposed scheme focuses fusion of concrete abstract features. It constructs detector (MsfNet) network considers relationship between single local environment. Combining global information with information, pyramid is constructed by fusing different depth layers in network. In addition, this also new extraction (CourNet), through way...

In order to enhance fault diagnosis precision, the wavelet packet analysis and probabilistic neural networks (PNN) are combined effectively. First, by selecting proper parameters, power spectrum of signals decomposed analysis, which predigests choosing method eigenvectors. Second, a based on PNN is presented. The uses Bayesian classifying decision making theory constitute mathematic model system, with Gauss function as activating function. possesses characteristics strong nonlinear...

For electronic commerce, security is the important problem. SSL protocol becomes supreme available because of its simplicity and credibility. The hidden trouble in application decided by limiting factors itself American restrictions on cipher exporting. This paper to offer proposals improve based analysis it. On basis feasibility, proxy designed installed at both client side server as transition software, so ensure commerce. A designing modularize adopted guarantee extensibility flexibility...

An improved VPN (virtual private network) system based on SSL (secure socket layer) protocol is discussed to overwhelm the defect of traditional VPN, which required install client software and do a complex operation. The concept critical technique are discussed, then architecture working process handshake layer record analyzed in detail because they very important for ensuring network safety SSL. Based above technology, an designed; it overcomes flaw currently enhances security system. In...

When neural networks are used to forecast short-term power load, it can learn the experience by training and generate mapping rules, but these rules not directly understood in network. By using method of integrating fuzzy logic, only settle historical load information. Moreover, logic considers factors which have great effect varying, such as air temperature holidays, etc. According own characteristics membership function constructed, modifying basic heft is realized, enhance forecasting...

A fault diagnosis method based on stacked denosing autoencoder network is proposed to apply deep learning equipment diagnosis. deep-layer model established, pre-training performed in a layer-by-layer greedy coding mode achieve adaptive extraction and mining of high-dimensional features, then back-propagation algorithm used supervise fine-tune the model. The integrates two steps feature state classification, gets rid dependence traditional machine methods artificially extracted sample...

According to the non-linear relation characteristic of load, a short-term load forecasting model based on fuzzy neural networks was presented. In model, inference and defuzzification were completed by networks, weight values given definite knowledge meaning. The membership function layer selected translate input variables into variables. Then new algorithm discussed finish inference. Finally, obtained proper defuzzification. simulation results show preferable capability model.

In fault diagnosis practice of recent years, the neural networks obtain many harvests, but it has a lot questions in network structure selecting and training. paper, power spectrum signals are decomposed by wavelet analysis, which predigests choosing method eigenvectors, model based on least squares support vector machine (LSSVM) is presented. Via structural risk minimization principle to enhance extensive ability, preferably solves practical problems, such as small sample, non-linear, high...

It is very important to select input variables when the support vector machine forecasting model proposed. The selection for short-term load relevant performance of forecasting. By using correlation coefficient idea on forecasting, a systemic and operable method sets first An example given. result shows that more preferable set can be obtained, errors are smaller, which validates effective

Clustering is a hot research field in data mining. There are so many methods or algorithms designed for different type set on which analysis action operates. Local Agglomerative Characteristic (LAC) based Algorithm, this paper, presented clustering, can handle clusters of size, shapes, and densities, work well distributed natural variant set. The new algorithm design the survey Jarvis-Patrick Shared Nearest Neighbor (SNN) density-based clustering algorithm, then deal with these kind avoid...

A vowel is a sound where air coming from the lungs not blocked by mouth or throat. The articulatory features that distinguish different sounds are said to determine vowel's quality. It very important recognize classes. recognition model based on improved least squares support vector machine (LSSVM) presented. In model, when LSSVM used in recognition, it presented choose parameter of kernel function dynamic, which enhances preciseness rate diagnosis. Fibonacci symmetry searching algorithm...