Recently, contrastive learning has been shown to be effective in improving pre-trained language models (PLM) derive high-quality sentence representations. It aims pull close positive examples enhance the alignment while push apart irrelevant negatives for uniformity of whole representation space.However, previous works mostly adopt in-batch or sample from training data at random. Such a way may cause sampling bias that improper (false and anisotropy representations) are used learn...

Extracting buildings automatically from high-resolution aerial images is a significant and fundamental task for various practical applications, such as land-use statistics urban planning. Recently, methods based on deep learning, especially the fully convolution networks, achieve impressive scores in this challenging semantic segmentation task. However, lack of global contextual information careless upsampling method limit further improvement performance building extraction To simultaneously...

Deep neural networks have performed well on various benchmark vision tasks with large-scale labeled training data; however, such data is expensive and time-consuming to obtain. Due domain shift or dataset bias, directly transferring models trained a source another sparsely unlabeled target often results in poor performance. In this paper, we consider the adaptation problem image emotion recognition. Specifically, study how adapt discrete probability distributions of emotions from an...

The accurate forecasting of urban taxi demands, which is a hot topic in intelligent transportation research, challenging due to the complicated spatial-temporal dependencies, dynamic nature, and uncertainty traffic. To make full use global local correlations between traffic flows on road sections, this paper presents deep learning model based graph convolutional network, long short-term memory (LSTM), multitask learning. First, an undirected was formed by considering spatial pattern...

Most real-time semantic segmentation networks use shallow architectures to achieve fast inference speeds. This approach, however, limits a network’s receptive field. Concurrently, feature information extraction is restricted single scale, which reduces the ability generalize and maintain robustness. Furthermore, loss of image spatial details negatively impacts accuracy. To address these limitations, this paper proposes Multiscale Context Pyramid Pooling Spatial Detail Enhancement Network...

Natural images exhibit a high degree of complexity, randomness and irregularity in color texture, however fractal can be an effective tool to describe various irregular phenomena nature. Fractal dimensions are important because they defined connection with real-world data, measured approximately by means experiments. In this paper, we proposed dimension estimation method for RGB images. the method, present hyper-surface partition which considers as continuous divide image into nonoverlapped...

The analysis of fractal patterns has grown during the past years, mainly due to wide range applications diverse scientific areas where fractals have been explored. It turns out that key tool study complexity a given system is Fractal Dimension (FD), since this its main invariant which throws quite useful information about it presents when being examined with enough level detail. In proposed method, we adopt hyper-surface partition method considers as continuous and divides image into...

In this paper, we study the problem of monocular 3D human pose estimation based on deep learning. Due to single view limitations, cannot avoid inherent occlusion problem. The common methods use multi-view method solve However, single-view images be used directly in methods, which greatly limits practical applications. To address above-mentioned issues, propose a novel end-to-end network for estimation. First, generator predict 2D poses from view. Secondly, simple but effective data...

Structures and functions of proteins play various essential roles in biological processes. The newly discovered can be predicted by comparing their structures with that known-functional proteins. Many approaches have been proposed for measuring the protein structure similarity, such as template-modeling (TM)-score method, GRaphlet (GR)-Align method well commonly used root-mean-square deviation (RMSD) measures. However, alignment comparisons between similarity cost much time on large dataset,...

Breast cancer is one of the most common cancers affecting women lives worldwide. It usually quite difficult for radiologists to accurately distinguish between malignant and benign tumor in digital mammograms. An intelligent classifier based on conventional machine learning algorithms can help classifying abnormal breast mass diagnosing cancer. Recently, deep has attracted much research attention medical image analysis due its higher accuracy capability features from annotated imaging data...

In this paper, the Adaboost algorithm is optimised to classify and predict user's credit risk by combining long short term memory neural network LSTM. The dataset was firstly divided, transposed, normalised, tiled format converted then model trained tested. During training process, it observed that loss on set gradually decreases optimally fits data converges optimal solution. confusion matrix shows of 2914 customers correctly predicted in with an accuracy 83.3%. performs well able...

Abstract Ships navigating in the water need to consider their safety first, and abnormal sinking amount of ship will be a great threat navigation ship. In this paper, relationship between hydrodynamic parameters is studied order realize reasonable control by setting different so as ensure navigation. Through Taylor’s formula calculate navigation, through combination potential flow theory boundary conditions, finally using Green’s function method surface element method, it calculated that...

The stability of a colluvial slope, which is different from rock or soil determined by the properties both bedrock and colluvium. Coupled with artificial excavation environmental effects, factors such slopes are complicated. To rapidly effectively eva luate risk cutting evaluation system for this type slope established herein. First, an index established, reasonable indices selected. Second, fuzzy analytic hierarchy process (FAHP) applied, pairwise comparison matrix, that must satisfy...

Supplier selection is a multiple-criteria decision-making (MCDM) problem. However, selecting proper supplier not an easy decision because lots of evaluating criteria such as product quality, cost, delivery dependability, customer service and so on should be considered. This paper aimed to present intuitionistic fuzzy approach deal with the problem in supply chain system. First, we analyzed factors for based some related literature. Furthermore, proposed new TOPSIS Finally, example was used...