To construct small mobile networks without performance loss and address the over-fitting issues caused by less abundant training datasets, this paper proposes a novel super sparse convolutional (SSC) kernel, its corresponding network is called SSC-Net. In SSC every spatial kernel has only one non-zero parameter these positions are all different. The can effectively select pixels from feature maps according to perform on them. Therefore, preserve general characteristics of geometric channels’...

The escalation of industrialization has worsened air quality, underscoring the essential need for accurate forecasting to inform policies and protect public health. Current research primarily emphasized individual spatiotemporal features prediction, neglecting interconnections between these features. To address this, we proposed generative Comprehensive Scale Spatiotemporal Fusion Air Quality Predictor (CSST-AQP). novel dual-branch architecture combines multi-scale spatial correlation...

Procedure planning in instructional videos, producing a structured and plannable action sequence facilitating the transition from start to goal states, has achieved significant progress. The dominant single-branch non-autoregressive paradigm guides generation through labels, overlooking limitation of absence intermediate visual information. Hence, we introduce procedure knowledge decoupled distillation strategy address above issue. This innovative deliberately lets teacher model see real...

Color image recognition based on real-number neural networks can encounter challenges such as redundant self-information and insufficient mutual information in ternary color feature extraction. These issues lead to inaccurate perception of by the model. Furthermore, deep with over 100 layers may experience a decline model performance due large number parameters. To overcome these challenges, this study extends into quaternion domain construct series models residual networks. The designs...

High-resolution fingerprint recognition has been a hot topic for many years. Compared with traditional image, high-resolution image can provide more features, such as pores and ridge contours. Introducing these features into comparison improve the accuracy reduce risk of identification errors. This paper proposes novel method comparing on images. The be divided two steps. In first step, fingerprints are aligned using pixel-category-distance-based data-driven descending algorithm....

In this paper, we propose a method based on the SVM algorithm to recognize dynamic hand gestures. The information of motion trajectory is captured by leap in three-dimension space. A new methodology feature extracting proposed guarantee length samples being same. elements vectors are ranged according two different criteria: one amplitude variation orientation angles, and other criterion order appearance features. Experimental results show that can classify gestures effectively.

In this paper, we proposed an effective method which can recognize dynamic hand gesture by analyzing the information of motion trajectory captured leap in three-dimension space. A simple spotting is tried. And orientation characteristics are quantified and coded as feature after pre-processing data. Then improved discrete HMM algorithm utilized to model classify gestures. Experimental results on a self-built database gestures (numbers 0-9) demonstrate effectiveness method.

With the increased model size of convolutional neural networks (CNNs), overfitting has become main bottleneck to further improve performance networks. Currently, weighting regularization methods have been proposed address problem and they perform satisfactorily. Since these cannot be used in all are usually not flexible enough different phases training test processes, this article proposes a multiscale conditional (MSC) method. MSC divides intermediate features into scales then generates new...

Pore is widely used because of its strong security and usefulness for live fingerprint detection recognition. There are considerable pores in a high-resolution image that can be However, the quality has become one bottlenecks method pore matching currently. In order to improve accuracy stability existing methods, this paper proposes enhancement technique based on neural network. The residual structure learn local features reconstruct original image. Experimental results indicate approach...

Real-time performance can be greatly improved, if the early recognition is implemented. In this paper, a dynamic hand gesture system proposed. The recognize before it completed. Our method based on Hidden Semi-Markov Models. Three-dimensional information of trajectory collected by leapmotion main data we used. Experiments dataset which established demonstrate effectiveness our method.

High-resolution fingerprint images contain three levels of features. Pores, as one the level 3 features, have wide attention due to its significant contribution recognition accuracy. An accurate and stable pore detection algorithm plays a key role on pore-based system. This paper proposes method for high-resolution images. The uses fully convolutional network combined with focal loss shortcut structure detect pores. proposed is tested database. Experimental results show that our outperforms...

High‐resolution fingerprint identification system (HRFIS) has become a hot topic in the field of academic research. Compared to traditional automatic system, HRFIS reduces risk being faked by using level 3 features, such as pores, which cannot be detected lower resolution images. However, there is serious problem HRFIS: are hundreds sweat pores one image, will spend considerable amount time for direct matching. The authors propose method match two images based on deterministic annealing...

Pore-based fingerprint recognition has been researched for decades. Many algorithms have proposed to improve the accuracy of system. However, accuracies are always improved at cost speed. This article proposes a novel method compare pores in high-resolution images using popular coarse-to-fine strategy. A multiple spatial pairwise local co-occurrence descriptor is calculation similarities between pores. It calculates statistics each pore its neighbors. The can establish correspondences more...

In order to address the overfitting problem caused by small or simple training datasets and large model’s size in Convolutional Neural Networks (CNNs), a novel Auto Adaptive Regularization (AAR) method is proposed this paper. The relevant networks can be called AAR-CNNs. AAR first using “abstraction extent” (predicted AE net) tiny learnable module (SE auto adaptively predict more accurate individualized regularization information. directly inserted into every stage of any popular trained end...

Pore-based fingerprint recognition has become more attractive in recent years because of its uniqueness and difficulty forgeability. However, most the existing methods use pore features for verification rather than indexing, them only local pores to evaluate similarities between fingerprints. In this article, we present a hierarchical pore-based high-resolution indexing system using two kinds (including global features). Our work consists major parts. First, design united model named deep...