This paper focuses on network pruning for image retrieval acceleration. Prevailing works target at the discriminative feature learning, while little attention is paid to how accelerate model inference, which should be taken into consideration in real-world practice. The challenge of models that middle-level preserved as much possible. Such different requirements and classification make traditional methods not suitable our task. To solve problem, we propose a new Progressive Local Filter...

In many real-world applications, data are represented by high-dimensional features. Despite the simplicity, existing K-means subspace clustering algorithms often employ eigenvalue decomposition to generate an approximate solution, which makes model less efficiency. Besides, their loss functions either sensitive outliers or small errors. this paper, we propose a fast adaptive (FAKM) type model, where function is designed provide flexible cluster indicator calculation mechanism, thereby...

Deep learning has shown significant successes in person reidentification (re-id) tasks. However, most existing works focus on discriminative feature and impose complex neural networks, suffering from low inference efficiency. In fact, extraction time is also crucial for real-world applications lightweight models are needed. Prevailing pruning methods usually pay attention to compact classification models. these suboptimal compacting re-id models, which produce continuous features sensitive...

Introduction In recent years, Unmanned Aerial Vehicles (UAVs) have increasingly been deployed in various applications such as autonomous navigation, surveillance, and object detection. Traditional methods for UAV navigation detection often relied on either handcrafted features or unimodal deep learning approaches. While these seen some success, they frequently encounter limitations dynamic environments, where robustness computational efficiency become critical real-time performance....

In pattern recognition and data mining, clustering is a classical technique to group matters of interest has been widely employed numerous applications. Among various algorithms, K-means (KM) most popular for its simplicity efficiency. However, with the rapid development social network, high-dimensional are frequently generated, which poses considerable challenge traditional KM as curse dimensionality. such scenarios, it difficult directly cluster that always contain redundant features...

The task of complex scene semantic segmentation is to classify and label the image pixel by pixel. For information in autonomous driving scenes, its characteristics such as many kinds targets various changes make more difficult, making FCN-based networks unable restore well. In contrast, encoder–decoder network structure represented SegNet UNet uses jump connections other methods information. Still, extraction shallow details simple unfocused. this paper, we propose a U-shaped convolutional...

This work explores the combination of LIBS technology and K-ELM algorithm for quantitative analysis total iron (TFe) content alkalinity sinter.

The aims of this study were to explore the feasibility using LIBS technology combined with a hybrid RF algorithm for quantitative analysis K in potash ore.

Architectural image segmentation refers to the extraction of architectural objects from remote sensing images. At present, most neural networks ignore relationship between feature information, and there are problems such as model overfitting gradient explosion. Thus, this paper proposes an improved UNet based on ResNet34 Attention Module (ResAt-UNet) solve related problems. The algorithm adds a two-layer residual structure (BasicBlock) regional enhancement attention mechanism (Space...

As CNNs are widely used in fields such as image classification and target detection, the total number of parameters computation models is gradually increasing. In addition, requirements on hardware resources power consumption for deploying becoming higher higher, leading to CNN being restricted certain specific platforms miniaturization practicality. Therefore, this paper proposes a convolutional-neural-network-processor design with an FPGA-based resource-multiplexing architecture, aiming...

River segmentation of remote sensing images is important research significance and application value for environmental monitoring, disaster warning, agricultural planning in an area. In this study, we propose a river model based on composite attention network to solve the problems abundant details interference non‐river information including bridges, shadows, roads. To improve efficiency, mechanism firstly introduced central region obtain global feature dependence information. Next,...

Abstract Urban laser radar point cloud building extraction is a hot spot in recent years, but the accurate distinction between vegetation, buildings and man-made objects has always been difficult point. In this paper, classification algorithm based on ICSF weakly correlated random forest are proposed for problem of low accuracy. Firstly, data ground-filtered by algorithm, then decision tree constructed, correlation analysis performed maximum mutual information coefficient. A with smallest...

In recent years, segmentation-based two-stage neural networks have emerged in the defect detection domain. Although these shown great success surface detection, they suffer from blurred edges, noise, and imbalanced data, which degrade performance. this study, we address limitations by presenting a new framework. To solve problem of with propose leveraging multilevel representations during segmentation. Furthermore, employ joint loss function that combines binary cross-entropy Dice to reduce...

<title>Abstract</title> Environmental perception and object detection are pivotalresearch topics in the marine domain. The sea surface presents unique challenges, including harsh weather conditions, wave interference, multi-scale targets, often resulting suboptimal results. To address these issues, we present an innovative solution: integrating Efficient Scale Fusion Module (ESFM) into advanced YOLO architecture, enhanced model, YOLO-ESFM. ESFM serves as both backbone head of network,...

Abstract To address the problems of low efficiency, large error and high bit rate in phase unwrapping high-frequency fringes by traditional time-phase method, this paper we propose a coding method that quantizes multivariate gray code domain. Instead embedding stepped into sinusoidal pattern, embed pattern which reduces levels to larger extent widens longitudinal width between each step pattern. After camera captures deformed is dequantified difference level, high-quality ladder word...