Edge-AI-Driven Framework with Efficient Mobile Network Design for Facial Expression Recognition

Wu, Yirui; Zhang, Lilai; Gu, Zonghua; Lu, Hanqing; Wan, Shaohua

doi:10.1145/3587038

Cited by 51 publications

(11 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We have made some analysis of the spearman correlation between the input and output of the model, but it may be possible to further study it through the interpretive ML algorithm in the future ( Jones et al., 2022 ). Moreover, in recent years, graph convolutional network (GCN), knowledge distillation (KD), edge artificial intelligence algorithms and other technologies have shown good potential in unsupervised learning ( Gou et al., 2022 ; Wu et al., 2023 ; Zhang et al., 2023 ), but how to apply these methods to the agricultural field remains to be further studied.…”

Section: Discussionmentioning

confidence: 99%

Simulating soil salinity dynamics, cotton yield and evapotranspiration under drip irrigation by ensemble machine learning

et al. 2023

View full text Add to dashboard Cite

Cotton is widely used in textile, decoration, and industry, but it is also threatened by soil salinization. Drip irrigation plays an important role in improving water and fertilization utilization efficiency and ensuring crop production in arid areas. Accurate prediction of soil salinity and crop evapotranspiration under drip irrigation is essential to guide water management practices in arid and saline areas. However, traditional hydrological models such as Hydrus require more variety of input parameters and user expertise, which limits its application in practice, and machine learning (ML) provides a potential alternative. Based on a global dataset collected from 134 pieces of literature, we proposed a method to comprehensively simulate soil salinity, evapotranspiration (ET) and cotton yield. Results showed that it was recommended to predict soil salinity, crop evapotranspiration and cotton yield based on soil data (bulk density), meteorological factors, irrigation data and other data. Among them, meteorological factors include annual average temperature, total precipitation, year. Irrigation data include salinity in irrigation water, soil matric potential and irrigation water volume, while other data include soil depth, distance from dripper, days after sowing (for EC and soil salinity), fertilization rate (for yield and ET). The accuracy of the model has reached a satisfactory level, R2 in 0.78-0.99. The performance of stacking ensemble ML was better than that of a single model, i.e., gradient boosting decision tree (GBDT); random forest (RF); extreme gradient boosting regression (XGBR), with R2 increased by 0.02%-19.31%. In all input combinations, other data have a greater impact on the model accuracy, while the RMSE of the S1 scenario (input without meteorological factors) without meteorological data has little difference, which is -34.22%~19.20% higher than that of full input. Given the wide application of drip irrigation in cotton, we recommend the application of ensemble ML to predict soil salinity and crop evapotranspiration, thus serving as the basis for adjusting the irrigation schedule.

show abstract

Section: Discussionmentioning

confidence: 99%

Simulating soil salinity dynamics, cotton yield and evapotranspiration under drip irrigation by ensemble machine learning

et al. 2023

View full text Add to dashboard Cite

show abstract

“…At the same time, most existing detection methods adopt a centralized processing mode, which is inefficient. However, the cloud edge collaborative operation mode can process data information at the network edge, greatly improving the efficiency of system analysis, which also provides ideas for optimizing target detection methods [21,22].…”

Section: Related Researchmentioning

confidence: 99%

An operation site security detection method based on point cloud data and improved YOLO algorithm under the architecture of the power internet of things

Yang,

Wang,

Guo

et al. 2024

The Journal of Engineering

View full text Add to dashboard Cite

An operation site safety detection method based on point cloud data and improved YOLO algorithm under the power Internet of Things architecture is proposed to address the complex environment of power construction sites and the poor effectiveness of most existing object detection methods. Firstly, an operation site safety supervision system was designed based on the power Internet of Things architecture, and efficient image processing was achieved through cloud edge collaboration. Then, point cloud data and on‐site monitoring information are used on the edge side to extract the accessible area, ensuring that the target is located in a safe area. Finally, the YOLO algorithm is improved in the cloud by using clustering algorithms, network structure optimization, and other methods, and used to detect targets and determine whether their behaviour meets the safety requirements of the operation site. Based on the PyTorch deep learning framework, the proposed method was experimentally demonstrated, and the results showed that its average detection accuracy and time were 94.53% and 68 ms, respectively, providing technical support for achieving remote monitoring of power operation sites.

show abstract

“…Attention modules are commonly utilized for feature extraction in various applications. In the study by [38], two attention modules, namely arbitrary‐oriented spatial pooling (ASP) and scalable frequency pooling (SFP), are proposed to enhance feature extraction and improve classification accuracy. The work presented by [39] introduces context‐aware deformable transformers for end‐to‐end chest abnormality detection (CDT‐CAD), which involves the construction of an iterative context‐aware feature extractor.…”

Section: Feature Extractionmentioning

confidence: 99%

A survey of feature matching methods

Huang,

Guo,

Wang

et al. 2024

IET Image Processing

View full text Add to dashboard Cite

Feature matching plays a crucial role in computer vision, with applications in visual localization, simultaneous localization and mapping (SLAM), image stitching, and more. It establishes correspondences between sets of feature points from multiple images, enabling various tasks. Over the years, feature matching has witnessed significant development, with an increasing number of methods being applied. However, different methods exhibit different degrees of applicability in different scenarios and requirements due to their different rationales. To cope with these issues, a comprehensive analysis and comparison of matching methods are essential. Existing reviews often lack coverage of deep learning models and focus more on feature detection and description, neglecting the matching process. This survey investigates feature detection, description, and matching techniques within the feature‐based image‐matching pipeline. Representative methods, their mechanisms, and application scenarios are also briefly introduced. In addition, comprehensive evaluations of classical and state‐of‐the‐art methods are conducted through extensive experiments on representative datasets. Particularly, matching‐based applications are compared to fully demonstrate the advantages of the methods. Lastly, this survey highlights current problems and development directions in matching methods, serving as a reference for researchers in the field.

show abstract

Edge-AI-Driven Framework with Efficient Mobile Network Design for Facial Expression Recognition

Cited by 51 publications

References 19 publications

Simulating soil salinity dynamics, cotton yield and evapotranspiration under drip irrigation by ensemble machine learning

Simulating soil salinity dynamics, cotton yield and evapotranspiration under drip irrigation by ensemble machine learning

An operation site security detection method based on point cloud data and improved YOLO algorithm under the architecture of the power internet of things

A survey of feature matching methods

Contact Info

Product

Resources

About