Firas Gerges, M. Boufadel, E. Bou‐Zeid, H. Nassif, J. T. Wang
Solar Irradiance depicts the light energy produced by the Sun that hits the Earth. This energy is important for renewable energy generation and is intrinsically fluctuating. Forecasting solar irradiance is crucial for efficient solar energy generation and management. Work in the literature focused on the short-term prediction of solar irradiance, using meteorological data to forecast the irradiance for the next hours, days, or weeks. Facing climate change and the continuous increase of greenhouse gas emissions, particularly from the use of fossil fuels, the reliance on renewable energy sources, such as solar energy, is expanding. Consequently, governments and practitioners are calling for efficient long-term energy generation plans, which could enable 100% renewable-based electricity systems to match energy demand. In this paper, we aim to perform the long-term prediction of solar irradiance, by leveraging the downscaled climate simulations of Global Circulation Models (GCMs). We propose a novel Bayesian deep learning framework, named DeepSI (denoting Deep Solar Irradiance), that employs bidirectional long short-term memory autoencoders, prefixed to a transformer, with an uncertainty quantification component based on the Monte-Carlo dropout sampling technique. We use DeepSI to predict daily solar irradiance for three different locations within the United States. These locations include the Solar Star power station in California, Medford in New Jersey, and Farmers Branch in Texas. Experimental results showcase the suitability of DeepSI for predicting daily solar irradiance from the simulated climate data. We further use DeepSI with future climate simulations to produce long-term projections of daily solar irradiance, up to year 2099.
{"title":"Deep Learning-Enabled Prediction of Daily Solar Irradiance from Simulated Climate Data","authors":"Firas Gerges, M. Boufadel, E. Bou‐Zeid, H. Nassif, J. T. Wang","doi":"10.1145/3583788.3583803","DOIUrl":"https://doi.org/10.1145/3583788.3583803","url":null,"abstract":"Solar Irradiance depicts the light energy produced by the Sun that hits the Earth. This energy is important for renewable energy generation and is intrinsically fluctuating. Forecasting solar irradiance is crucial for efficient solar energy generation and management. Work in the literature focused on the short-term prediction of solar irradiance, using meteorological data to forecast the irradiance for the next hours, days, or weeks. Facing climate change and the continuous increase of greenhouse gas emissions, particularly from the use of fossil fuels, the reliance on renewable energy sources, such as solar energy, is expanding. Consequently, governments and practitioners are calling for efficient long-term energy generation plans, which could enable 100% renewable-based electricity systems to match energy demand. In this paper, we aim to perform the long-term prediction of solar irradiance, by leveraging the downscaled climate simulations of Global Circulation Models (GCMs). We propose a novel Bayesian deep learning framework, named DeepSI (denoting Deep Solar Irradiance), that employs bidirectional long short-term memory autoencoders, prefixed to a transformer, with an uncertainty quantification component based on the Monte-Carlo dropout sampling technique. We use DeepSI to predict daily solar irradiance for three different locations within the United States. These locations include the Solar Star power station in California, Medford in New Jersey, and Farmers Branch in Texas. Experimental results showcase the suitability of DeepSI for predicting daily solar irradiance from the simulated climate data. We further use DeepSI with future climate simulations to produce long-term projections of daily solar irradiance, up to year 2099.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115679973","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Printer source identification is an important means of document inspection and plays an important role in forensic identification. In the research of printer source recognition, traditional methods basically rely on specific characters to recognize printed documents, but the recognition of Chinese printed documents is usually difficult because there are few or no specific characters. In view of this situation, this paper proposes a text-independent printer source identification method, which uses a graphical model to model the timing relationship of the printer, and then extracts the timing characteristics of the printer, which belong to the text-independent printer. Internal features, so that the method can be recognized without relying on specific characters, and has achieved good experimental results. Experimental data show that the proposed method is very useful for the traceability of printed documents.
{"title":"Printer Source Identification Based on Graph Model","authors":"Rui-Li Tian, Ziqi Zhu","doi":"10.1145/3583788.3583806","DOIUrl":"https://doi.org/10.1145/3583788.3583806","url":null,"abstract":"Printer source identification is an important means of document inspection and plays an important role in forensic identification. In the research of printer source recognition, traditional methods basically rely on specific characters to recognize printed documents, but the recognition of Chinese printed documents is usually difficult because there are few or no specific characters. In view of this situation, this paper proposes a text-independent printer source identification method, which uses a graphical model to model the timing relationship of the printer, and then extracts the timing characteristics of the printer, which belong to the text-independent printer. Internal features, so that the method can be recognized without relying on specific characters, and has achieved good experimental results. Experimental data show that the proposed method is very useful for the traceability of printed documents.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124460824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Deep learning methods, including convolutional neural networks (CNNs) and recurrent neural network (RNN), have been used for analysis of brain network, e.g., dynamic functional connectivity (dFC) network. However, CNN usually extract local features of brain network, ignoring the temporal information of dFC network. In addition, diversity feature representations of brain network can be obtained using convolutional kernels with different scales, these representations may contain complementary information that could be used for further improving the diagnosis performance of brain disease (e.g., Alzheimer’s Disease, AD). To address this problem, in this paper, we propose a convolutional recurrent neural network with multi-scale kernels (MSK-CRNN) learning framework for brain disease classification with fMRI data. Specifically, we build a convolutional layer with multi-scale kernels to extract different-yet-complementary features from constructed dFC networks, and use a long short-term memory (LSTM) layer to further extract temporal information of dFC networks. The experimental results on 174 subjects with 563 scans from Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset demonstrate that, compared with the existing methods, the proposed MSK-CRNN method can further improve the performance of AD classification.
{"title":"Convolutional Recurrent Neural Network with Multi-Scale Kernels on Dynamic Connectivity Network for AD Classification","authors":"Xingyu Zhang, Biao Jie, Jianhui Wang","doi":"10.1145/3583788.3583798","DOIUrl":"https://doi.org/10.1145/3583788.3583798","url":null,"abstract":"Deep learning methods, including convolutional neural networks (CNNs) and recurrent neural network (RNN), have been used for analysis of brain network, e.g., dynamic functional connectivity (dFC) network. However, CNN usually extract local features of brain network, ignoring the temporal information of dFC network. In addition, diversity feature representations of brain network can be obtained using convolutional kernels with different scales, these representations may contain complementary information that could be used for further improving the diagnosis performance of brain disease (e.g., Alzheimer’s Disease, AD). To address this problem, in this paper, we propose a convolutional recurrent neural network with multi-scale kernels (MSK-CRNN) learning framework for brain disease classification with fMRI data. Specifically, we build a convolutional layer with multi-scale kernels to extract different-yet-complementary features from constructed dFC networks, and use a long short-term memory (LSTM) layer to further extract temporal information of dFC networks. The experimental results on 174 subjects with 563 scans from Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset demonstrate that, compared with the existing methods, the proposed MSK-CRNN method can further improve the performance of AD classification.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"267 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123405535","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Venkata Duvvuri, Gahyoung Lee, Yuwei Hsu, Asha Makwana, C. Morgan
Automatic Item Generation (AIG) is increasingly used to process large amounts of information and scale the demand for computerized testing. Recent work in Artificial Intelligence for AIG (aka Natural Question Generation-NQG), states that even newer AIG techniques are short in syntactic, semantic, and contextual relevance when evaluated qualitatively on small datasets. We confirm this deficiency quantitatively over large datasets. Additionally, we find that human evaluation by Subject Matter Experts (SMEs) conservatively rejects at least ∼9% portion of AI test questions in our experiment over large diverse dataset topics. Here we present an analytical study of these differences, and this motivates our two-phased post-processing AI daisy chain machine learning (ML) architecture for selection and editing of AI generated questions using current techniques. Finally, we identify and propose the first selection step in the daisy chain using ML with 97+% accuracy, and provide analytical guidance for development of the second editing step with a measured lower bound on a BLEU score improvement of 2.4+%.
{"title":"Post Processing Selection of Automatic Item Generation in Testing to Ensure Human-Like Quality with Machine Learning","authors":"Venkata Duvvuri, Gahyoung Lee, Yuwei Hsu, Asha Makwana, C. Morgan","doi":"10.1145/3583788.3583800","DOIUrl":"https://doi.org/10.1145/3583788.3583800","url":null,"abstract":"Automatic Item Generation (AIG) is increasingly used to process large amounts of information and scale the demand for computerized testing. Recent work in Artificial Intelligence for AIG (aka Natural Question Generation-NQG), states that even newer AIG techniques are short in syntactic, semantic, and contextual relevance when evaluated qualitatively on small datasets. We confirm this deficiency quantitatively over large datasets. Additionally, we find that human evaluation by Subject Matter Experts (SMEs) conservatively rejects at least ∼9% portion of AI test questions in our experiment over large diverse dataset topics. Here we present an analytical study of these differences, and this motivates our two-phased post-processing AI daisy chain machine learning (ML) architecture for selection and editing of AI generated questions using current techniques. Finally, we identify and propose the first selection step in the daisy chain using ML with 97+% accuracy, and provide analytical guidance for development of the second editing step with a measured lower bound on a BLEU score improvement of 2.4+%.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126757495","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In actual prediction scenarios, attribute features include time data, weather data, and energy consumption data. The relationship between attribute features is very complex. Exploring the relationship between feature attributes and decision sets with fast attribute reduction can reduce the amount of model training data. Since seasonal and temporal variations greatly influence weather and energy consumption data, the moving horizon method is used to update the features and improve the accuracy of energy consumption prediction.From above, an energy consumption prediction model based on fast attribute reduction of weighted neighborhood rough set with moving horizon long short-term memory neural network(LSTM) is proposed in this paper. In the experiment of predicting the actual energy consumption of a building, the model evaluation results show that 20% of training data is reduced at the expense of 0.4% classification accuracy. Compared with the traditional non-rolling method, the Root Mean Square Error (RMSE) of the moving horizon LSTM prediction method is reduced by 33.08% on average, and the training speed is increased by 5.25% on average. The prediction effect is better. Therefore, this prediction model can be used to predict building energy consumption quickly and accurately and has strong robustness and generalization ability, which provides the theoretical basis and method support for fine management of building energy consumption, building energy conservation, and emission reduction.
{"title":"Research on energy consumption prediction based on fast attribute reduction of weighted neighborhood rough set with moving horizon","authors":"Jun Tan, Qun Hou, Xin Liu, Yunke Xiong","doi":"10.1145/3583788.3583791","DOIUrl":"https://doi.org/10.1145/3583788.3583791","url":null,"abstract":"In actual prediction scenarios, attribute features include time data, weather data, and energy consumption data. The relationship between attribute features is very complex. Exploring the relationship between feature attributes and decision sets with fast attribute reduction can reduce the amount of model training data. Since seasonal and temporal variations greatly influence weather and energy consumption data, the moving horizon method is used to update the features and improve the accuracy of energy consumption prediction.From above, an energy consumption prediction model based on fast attribute reduction of weighted neighborhood rough set with moving horizon long short-term memory neural network(LSTM) is proposed in this paper. In the experiment of predicting the actual energy consumption of a building, the model evaluation results show that 20% of training data is reduced at the expense of 0.4% classification accuracy. Compared with the traditional non-rolling method, the Root Mean Square Error (RMSE) of the moving horizon LSTM prediction method is reduced by 33.08% on average, and the training speed is increased by 5.25% on average. The prediction effect is better. Therefore, this prediction model can be used to predict building energy consumption quickly and accurately and has strong robustness and generalization ability, which provides the theoretical basis and method support for fine management of building energy consumption, building energy conservation, and emission reduction.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"95 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130407117","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
D. Jain, Akshit Khanna, Bhavya Gera, Dhiraj Sangwan
Object Detection is the task of detecting and localizing objects of importance in visual media. The rapid increase in the number of powerful end devices such as mobiles and surveillance cameras have to lead to increasing in both resources and media generation. Object Detection has now become possible on end devices, but certain challenges need to be tackled to fully utilize the resources. Federated Learning is one such framework that leverages the end device resources to build machine learning models while preserving data privacy. We model the federated learning framework for object detection on real-world heterogeneous datasets using a novel dampened harmonic optimizer to enhance local learning on the end device and hence reducing the communication cost during the learning process. We provide comparison of the commonly used FedAvg with our FedHarm optimization on multiple object detection models and datasets to demonstrate the merits of our proposition. Our method FedHarm with its dampened updates allows for greater local computation which reduces the overall communication rounds between edge devices and cloud and better handles heterogeneity in real-world datasets. FedHarm leads to faster convergence by 41% on average over FedAvg which is supported by extensive experiments.
{"title":"Federated Learning based Object Detection using Dampened Harmonic Optimization","authors":"D. Jain, Akshit Khanna, Bhavya Gera, Dhiraj Sangwan","doi":"10.1145/3583788.3583815","DOIUrl":"https://doi.org/10.1145/3583788.3583815","url":null,"abstract":"Object Detection is the task of detecting and localizing objects of importance in visual media. The rapid increase in the number of powerful end devices such as mobiles and surveillance cameras have to lead to increasing in both resources and media generation. Object Detection has now become possible on end devices, but certain challenges need to be tackled to fully utilize the resources. Federated Learning is one such framework that leverages the end device resources to build machine learning models while preserving data privacy. We model the federated learning framework for object detection on real-world heterogeneous datasets using a novel dampened harmonic optimizer to enhance local learning on the end device and hence reducing the communication cost during the learning process. We provide comparison of the commonly used FedAvg with our FedHarm optimization on multiple object detection models and datasets to demonstrate the merits of our proposition. Our method FedHarm with its dampened updates allows for greater local computation which reduces the overall communication rounds between edge devices and cloud and better handles heterogeneity in real-world datasets. FedHarm leads to faster convergence by 41% on average over FedAvg which is supported by extensive experiments.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"56 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123823882","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Most existing multimodal multi-objective evolutionary algorithms only search the global Pareto front of the problem while ignoring the excellent local Pareto front of the problem. To address this issue, an optimization algorithm with population clustering mechanism is proposed to settle multimodal multi-objective problems with local Pareto front. At the first step, a partitioning method is used to divide the total population into main rank and other ranks and a population clustering method is proposed to repartition the entire population into global Pareto front subpopulations and local Pareto front subpopulations. In the second step, each subpopulation evolves independently and the diversity in the objective space and decision space are considered simultaneously. An improved density adaptive adjustment strategy is put forward to enhance the diversity of the population in the decision space. In the experimental part, the algorithm is compared with several other state-of-the-art algorithms using the CEC 2019 MMOPs test case, and the result of the experiment confirm that the algorithm proposed shows excellent performance.
{"title":"Solving Multimodal Multi-Objective Problems with Local Pareto Front using a Population Clustering Mechanism","authors":"Fan Li, Kai Zhang, Chaonan Shen, Zhiwei Xu","doi":"10.1145/3583788.3583793","DOIUrl":"https://doi.org/10.1145/3583788.3583793","url":null,"abstract":"Most existing multimodal multi-objective evolutionary algorithms only search the global Pareto front of the problem while ignoring the excellent local Pareto front of the problem. To address this issue, an optimization algorithm with population clustering mechanism is proposed to settle multimodal multi-objective problems with local Pareto front. At the first step, a partitioning method is used to divide the total population into main rank and other ranks and a population clustering method is proposed to repartition the entire population into global Pareto front subpopulations and local Pareto front subpopulations. In the second step, each subpopulation evolves independently and the diversity in the objective space and decision space are considered simultaneously. An improved density adaptive adjustment strategy is put forward to enhance the diversity of the population in the decision space. In the experimental part, the algorithm is compared with several other state-of-the-art algorithms using the CEC 2019 MMOPs test case, and the result of the experiment confirm that the algorithm proposed shows excellent performance.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"68 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124516113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In dealing with the large discrete action spaces, a joint action representation and prioritized experience replay method is proposed in this paper, which consists of three modules. In the first module, we use the k-nearest neighbor method to reduce the dimensionality of the original action space, generating a compact action space, and then the critic network is introduced to further evaluate and filter this compact space to obtain the optimal action. Note that the optimal action may have inconsistency with the actual desired action. Then in the second module, we introduce a multi-step update technique to reduce the training variance when storing data in the replay buffer. In the third module, considering the existence of correlation between samples when sampling data, we assign the corresponding weight to the sample experience by calculating the absolute value of temporal difference error and use such a non-uniform sampling method to prioritize the samples for sampling. Experimental results on four benchmark environments demonstrate the effectiveness and efficiency of the proposed method in dealing with the large discrete action spaces.
{"title":"Joint Action Representation and Prioritized Experience Replay for Reinforcement Learning in Large Discrete Action Spaces","authors":"Xueyu Wei, Wei Xue, Wei Zhao, Yuanxia Shen, Gaohang Yu","doi":"10.1145/3583788.3583802","DOIUrl":"https://doi.org/10.1145/3583788.3583802","url":null,"abstract":"In dealing with the large discrete action spaces, a joint action representation and prioritized experience replay method is proposed in this paper, which consists of three modules. In the first module, we use the k-nearest neighbor method to reduce the dimensionality of the original action space, generating a compact action space, and then the critic network is introduced to further evaluate and filter this compact space to obtain the optimal action. Note that the optimal action may have inconsistency with the actual desired action. Then in the second module, we introduce a multi-step update technique to reduce the training variance when storing data in the replay buffer. In the third module, considering the existence of correlation between samples when sampling data, we assign the corresponding weight to the sample experience by calculating the absolute value of temporal difference error and use such a non-uniform sampling method to prioritize the samples for sampling. Experimental results on four benchmark environments demonstrate the effectiveness and efficiency of the proposed method in dealing with the large discrete action spaces.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114644627","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the context of the normalization of the COVID-19, wearing a mask is an effective way to prevent the spread of the COVID-19. It is an important but highly challenging task to detect people not wearing masks in crowded places in time. Automatic mask wearing detection based on monitored images has also become a current research hotspot. This paper proposes a model to detect the masked face by utilizing RetinaFace, which uses Res2Net as the backbone network, and enhances feature extraction by introducing a weighted bidirectional feature pyramid and CBAM (Convolutional Block Attention Module). Comparative experiments are done based on the actual scene, The experimental results demonstrate that the Retinaface_Mask has achieved better detection results than the Retinaface. The mean average precision of Retinaface_Mask reaches 86.92%, compared with the Retinaface, it is improved by 1.43 percentage points.
{"title":"A Mask Detection Algorithm Based on RetinaFace","authors":"Qingqing Huang, Wei Pan, Xing Fan","doi":"10.1145/3583788.3583818","DOIUrl":"https://doi.org/10.1145/3583788.3583818","url":null,"abstract":"In the context of the normalization of the COVID-19, wearing a mask is an effective way to prevent the spread of the COVID-19. It is an important but highly challenging task to detect people not wearing masks in crowded places in time. Automatic mask wearing detection based on monitored images has also become a current research hotspot. This paper proposes a model to detect the masked face by utilizing RetinaFace, which uses Res2Net as the backbone network, and enhances feature extraction by introducing a weighted bidirectional feature pyramid and CBAM (Convolutional Block Attention Module). Comparative experiments are done based on the actual scene, The experimental results demonstrate that the Retinaface_Mask has achieved better detection results than the Retinaface. The mean average precision of Retinaface_Mask reaches 86.92%, compared with the Retinaface, it is improved by 1.43 percentage points.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124463032","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the problem of calculating hull elements using the table of offsets, the sparsity between hull slices will bring uncertainty and error to the calculation. Therefore, this paper proposes a refinement algorithm of the ship hull based on the table of offsets: Firstly, the NURBS curve for the hull is constructed based on the table of offsets, and the hull's NURBS surface is obtained through the skinning algorithm. Secondly, the IR-BFS algorithm is used to inverse the knot parameters of the stations of the target station in the hull's NURBS surface. Thirdly, based on the knot parameters and the hull NURBS surface expression, the hull section, after refinement of the target station, is obtained. In constructing the hull's NURBS surface, the hull section is first expressed using the NURBS interpolation algorithm and the flattening algorithm of the NURBS based on the IR-BFS algorithm. Then the skinning algorithm is improved by fixing the -direction knot parameters to express the expressed hull NURBS cross-section as a hull's NURBS surface, which improves the computational efficiency. The effectiveness of the improved skinning algorithm is judged by comparing the increase in the number of control points and the computational time consumption in the expression of the hull NURBS surface before and after the improved skinning algorithm. The usability of the refinement algorithm of the hull section is verified by comparing the hull section based on the table of offsets with the refined hull section. The experimental results show that the improved skinning algorithm can effectively improve the speed of NURBS surface generation; The proposed refinement algorithm of the hull section can effectively generate refined sections through refinement intervals.
{"title":"Fast NURBS Skinning Algorithm and Ship Hull Section Refinement Model","authors":"Kaige Zhu, Guoyou Shi, Jiao Liu, Jiahui Shi, Yuchuang Wang, Xing Jiang","doi":"10.1145/3583788.3583792","DOIUrl":"https://doi.org/10.1145/3583788.3583792","url":null,"abstract":"In the problem of calculating hull elements using the table of offsets, the sparsity between hull slices will bring uncertainty and error to the calculation. Therefore, this paper proposes a refinement algorithm of the ship hull based on the table of offsets: Firstly, the NURBS curve for the hull is constructed based on the table of offsets, and the hull's NURBS surface is obtained through the skinning algorithm. Secondly, the IR-BFS algorithm is used to inverse the knot parameters of the stations of the target station in the hull's NURBS surface. Thirdly, based on the knot parameters and the hull NURBS surface expression, the hull section, after refinement of the target station, is obtained. In constructing the hull's NURBS surface, the hull section is first expressed using the NURBS interpolation algorithm and the flattening algorithm of the NURBS based on the IR-BFS algorithm. Then the skinning algorithm is improved by fixing the -direction knot parameters to express the expressed hull NURBS cross-section as a hull's NURBS surface, which improves the computational efficiency. The effectiveness of the improved skinning algorithm is judged by comparing the increase in the number of control points and the computational time consumption in the expression of the hull NURBS surface before and after the improved skinning algorithm. The usability of the refinement algorithm of the hull section is verified by comparing the hull section based on the table of offsets with the refined hull section. The experimental results show that the improved skinning algorithm can effectively improve the speed of NURBS surface generation; The proposed refinement algorithm of the hull section can effectively generate refined sections through refinement intervals.","PeriodicalId":292167,"journal":{"name":"Proceedings of the 2023 7th International Conference on Machine Learning and Soft Computing","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126453086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：