Machine learning (ML) offers considerable promise for the design of new molecules and materials. In real-world applications, problem is often domain-specific, suffers from insufficient data, particularly labeled ML training. this study, we report a data-efficient, deep-learning framework molecular discovery that integrates coarse-grained functional-group representation with self-attention mechanism to capture intricate chemical interactions. Our approach exploits group-contribution theory...

Particle filter is a probability estimation method based on Bayesian framework and it has unique advantage to describe the target tracking non-linear non-Gaussian. In this paper, Firstly, analyses particle degeneracy sample impoverishment in multi-target algorithm, secondly, applies Markov Chain Monte Carlo (MCMC) improve re-sampling process enhance performance of algorithm. Finally, proposed certificated by experiment that multiple targets similar appearance complex motion. The results show...

Purpose To overcome the shortcomings of traditional dynamic parameter identification methods in accuracy and efficiency, this paper aims to propose a method based on improved iterative reweighted least squares (IIRLS) algorithm. Design/methodology/approach First, Newton–Euler is used establish model robot, which linearized reorganized. Then, taking Fourier series as excitation trajectory, optimization with objective function established optimized. manipulator runs optimized trajectory...

With the rapid growth of China's transportation, traffic accidents were in a growing trend year by year, and tracking hit-and-run vehicles had caught hotspot people concern. Therefore, how to reduce road improve safety level become important issues that need be resolved. This paper whose research objects are is based on intelligent accident treatment system. After researching application Mean shift Kalman filter field target tracking, this presented an effective method which combined...

Aiming at the problems of low segmentation accuracy noise image, poor immunity existing models and adaptability to complex environment, a image algorithm using anisotropic diffusion nonconvex functional was proposed. First, focusing on “staircase effect”, introduced into energy model for smooth denoising. Second, validity were established by proving that there no global minimum in solution space model; improved then used obtain clear edge while maintaining integrity. Third, obtained from...

This text proposed the target segmentation algorithm that combined local energy information with improved signed distance regularization term. The adds energy, curve length constraint and term to global image of traditional C-V model. new inherits advantages functional adequately, accurately drives level set evolution contour. It effectively realized uneven color in less iteration. On other hand, avoids re-initialization function, increases computational efficiency, maintains stability...

Recent studies show that face recognition and verification in deep learning can achieve impressive performance aimed at the frontal face. Due to randomness of people's activities, faces are usually captured different views instead front. In practical application, most current methods become quite difficult dealing with tasks multi-view, thus these cases a better algorithm is required. this paper, we propose convolutional framework which combines SphereFace-20 Batch Normalization increases...

The energy functional of the CV and LBF model is single, which makes curve to get into local minimum easily during evolution process, results inaccurate segmentation images with nonuniform grayscale nonsmooth edges. proposed algorithm, based on entropy fitting under constraint nonconvex regularization term, used deal such problems. In this global information are fitted avoid falling optimum, term imported for protect edge smoothing. First, evolve approximate contour target segmentation....

The traditional mean shift algorithm used fixed kernels or symmetric kernel function, which will cause the target tracking lost failure. based on with adaptive bandwidth was proposed. Firstly, signed distance constraint function introduced to produce anisotropic function. This satisfies that value of region outside is zero, provides accurate window for tracking. Secondly, calculate center template, theory basis sum vector weights from sample point in zero. Thirdly, templates update...

The desire of looking for a more natural and efficient interactive method that human cantered are increasing recently. According to this status, paper try explore new human-computer system based on vision recognition, which conforms behaviour characteristic. This adopts RBF artificial neural network Hidden Markov model, locates user's input focus combined with head orientation the direction where finger pointed to. In way, recognition could be accuracy speed will improved. It has wide...

Propose a kind of electric transmission lines icing image segmentation algorithm based on quick bi-circulating level set to improve effect algorithm. First, carry out pre-treatment greying and decrease noise influence in image; second, fuzzy c mean value clustering propose SPKFCM by increasing space penalty function, which is used for automatic initialisation algorithm; finally, it shown that proposed has better efficiency experiment contrast image.

The proposed research work is aiming at the problem of enhancement in sharpness degraded images by using a deep learning method. A neural network introduced to study process images. construction training sample set done perform weight parameter on self-encoding which obtains and completes initial detection fuzzy region. Segmentation performed based threshold methods detect local blurred areas results are input multi-scale deblurring estimate residuals clear multiple scales. obtained...

To solve the problem of low edge protection index in traditional enhancement algorithm, a single image super-resolution algorithm based on deep learning is proposed. The feature extracted by sinusoidal two-dimensional transform function modulated Gaussian function. local Laplacian filter used to preprocess image, and method introduced enhance image. experimental results show that improved has higher index, can effectively improve accuracy, certain advantages.

In order to solve the problem that current vision system produces motion blur during imaging, a fast enhancement method for images based on improved deep learning is proposed. By detecting contour of image target, Wiener filtering restoration model established, combined with decompose gray tone function, and BNL-Means algorithm used calculate similarity between high-frequency blocks improve accuracy spatial feature extraction. Realize blurred images. Compared existing methods, it proved by...