Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687006
Chenguang Xu
The purposed of hyperspectral unmixing is to estimate the spectral signatures composing the data (endmembers) and their abundance fractions. However, most of the traditional sparse unmixing methods are effective in the case of high signal-to-noise ratio (SNR), but is not good in the case of high noise. In order to solve this problem, we innovatively integrates adaptive total variation (ATV) regularization into hyperspectral sparse unmixing and propose a new hyperspectal sparse unmixing model named adaptive total variation regularized for sparse unmixing (SU_ATV). The model can adaptively adjust the horizontal difference and vertical difference of TV, can better optimize the efficiency of TV to improve the anti-noise performance. The experimental results show that SU_ATV has good anti-noise performance to the sparse unmixing.
{"title":"Adaptive Total Variation Regularized for Hyperspectral Unmixing","authors":"Chenguang Xu","doi":"10.1109/PIC53636.2021.9687006","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687006","url":null,"abstract":"The purposed of hyperspectral unmixing is to estimate the spectral signatures composing the data (endmembers) and their abundance fractions. However, most of the traditional sparse unmixing methods are effective in the case of high signal-to-noise ratio (SNR), but is not good in the case of high noise. In order to solve this problem, we innovatively integrates adaptive total variation (ATV) regularization into hyperspectral sparse unmixing and propose a new hyperspectal sparse unmixing model named adaptive total variation regularized for sparse unmixing (SU_ATV). The model can adaptively adjust the horizontal difference and vertical difference of TV, can better optimize the efficiency of TV to improve the anti-noise performance. The experimental results show that SU_ATV has good anti-noise performance to the sparse unmixing.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116171026","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687009
Jian Dong, Li Zhang, Zilong Liu, Zhiwei Lin, Zhiming Cai
As it is difficult to classify and identify the actions caused by the distortion of radar signal during acquisition process, this paper obtains the feature value of action signal through preprocessing such as abnormal point removal and wavelet filtering, and obtains the signal fluctuation section of action through short-term power spectral density. In the eight classification experiment and the nine classification experiment, the accuracies of traditional Bayesian network, BP network and support vector machine (SVM) are no higher than 90.0% For the test set with too small samples and some distortion, even using GWO-SVM, the recognition rate is still less than 90%. Therefore, this paper improves the wolf swarm position vector in GWO algorithm, and optimizes the penalty function and function radius in SVM model. The experimental results of our method show that the accuracies of eight classification and nine classification experiments are 92.4% and 90.4% respectively, which are better than those of SVM and GWO-SVM.
{"title":"An Action Recognition Method Based on Radar Signal with Improved GWO-SVM Algorithm","authors":"Jian Dong, Li Zhang, Zilong Liu, Zhiwei Lin, Zhiming Cai","doi":"10.1109/PIC53636.2021.9687009","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687009","url":null,"abstract":"As it is difficult to classify and identify the actions caused by the distortion of radar signal during acquisition process, this paper obtains the feature value of action signal through preprocessing such as abnormal point removal and wavelet filtering, and obtains the signal fluctuation section of action through short-term power spectral density. In the eight classification experiment and the nine classification experiment, the accuracies of traditional Bayesian network, BP network and support vector machine (SVM) are no higher than 90.0% For the test set with too small samples and some distortion, even using GWO-SVM, the recognition rate is still less than 90%. Therefore, this paper improves the wolf swarm position vector in GWO algorithm, and optimizes the penalty function and function radius in SVM model. The experimental results of our method show that the accuracies of eight classification and nine classification experiments are 92.4% and 90.4% respectively, which are better than those of SVM and GWO-SVM.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130927693","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}
Blind guidance system has always been a research hotspot for years. Although there are many kinds of blind guidance systems on the market, most of them prompt from the perspective of a single sense of tactile or auditory. The blind guidance method of single sense can be unstable and it does not fully mobilize other general senses of the with vision impairment. This paper designs and implements a multi-sensory blind guidance system that provides tactile and auditory sensations by using the ORB-SLAM and YOLO techniques. Based on the RGB-D camera, the local obstacle avoidance system is realized at the tactile level through the point cloud filtering that feedback the results to the user through vibrating motors. The improved ORB-SLAM can generate a dense navigation map to implement a global obstacle avoidance system through the coordinate transformation. Real-time target detection and the YOLO-based prompt voice system is implemented at the auditory level. The system can detect the specific category and give the location of obstacles as real-time voice messages. The functions mentioned above are integrated and verified as a smart cane. Experimental results show that the position and category of the obstacles in the surrounding environment can be detected accurately in real-time through our system. By combining YOLO and ORB- SLAM, we can provide a piece of useful auxiliary equipment to the community of vision impairment and enable users to move about safely.
{"title":"A Multi-Sensory Blind Guidance System Based on YOLO and ORB-SLAM","authors":"Chufan Rui, Yichen Liu, Junru Shen, Zhaobin Li, Zaipeng Xie","doi":"10.1109/PIC53636.2021.9687018","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687018","url":null,"abstract":"Blind guidance system has always been a research hotspot for years. Although there are many kinds of blind guidance systems on the market, most of them prompt from the perspective of a single sense of tactile or auditory. The blind guidance method of single sense can be unstable and it does not fully mobilize other general senses of the with vision impairment. This paper designs and implements a multi-sensory blind guidance system that provides tactile and auditory sensations by using the ORB-SLAM and YOLO techniques. Based on the RGB-D camera, the local obstacle avoidance system is realized at the tactile level through the point cloud filtering that feedback the results to the user through vibrating motors. The improved ORB-SLAM can generate a dense navigation map to implement a global obstacle avoidance system through the coordinate transformation. Real-time target detection and the YOLO-based prompt voice system is implemented at the auditory level. The system can detect the specific category and give the location of obstacles as real-time voice messages. The functions mentioned above are integrated and verified as a smart cane. Experimental results show that the position and category of the obstacles in the surrounding environment can be detected accurately in real-time through our system. By combining YOLO and ORB- SLAM, we can provide a piece of useful auxiliary equipment to the community of vision impairment and enable users to move about safely.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127892733","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687090
Jianshe Liu
The existing water cannons are basically open-loop control, which can not provide real-time feedback on the position difference between the shooting flow point and the target, especially for the moving target, there is a large strike error. The jet closed-loop control technology is an important way to realize the accurate and continuous strike of water cannon on target. Considering only the influence of gravity and air resistance, a jet closed -loop control method is proposed. In this method, the horizontal Angle and pitch Angle of the water cannon are adjusted by geometric calculation and the jet motion trajectory model are established respectively. On the basis of image processing, a method to adjust the Angle of the water cannon again is designed, and the feasibility of this method is verified by a lot of simulation experiments. Experiments show that this method can dynamically adjust the jet Angle in real time according to the target bearing, and has high accuracy and real time.
{"title":"The Jet Closed-Loop Control Method Based on Image Processing","authors":"Jianshe Liu","doi":"10.1109/PIC53636.2021.9687090","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687090","url":null,"abstract":"The existing water cannons are basically open-loop control, which can not provide real-time feedback on the position difference between the shooting flow point and the target, especially for the moving target, there is a large strike error. The jet closed-loop control technology is an important way to realize the accurate and continuous strike of water cannon on target. Considering only the influence of gravity and air resistance, a jet closed -loop control method is proposed. In this method, the horizontal Angle and pitch Angle of the water cannon are adjusted by geometric calculation and the jet motion trajectory model are established respectively. On the basis of image processing, a method to adjust the Angle of the water cannon again is designed, and the feasibility of this method is verified by a lot of simulation experiments. Experiments show that this method can dynamically adjust the jet Angle in real time according to the target bearing, and has high accuracy and real time.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128843529","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687084
Chunzhi Hou, Jiarui Shi, Baohang Zhang
In recent years, the role of a single dendritic neural structures with non-linear localisation in computing has attracted a lot of attention from the industry. The dendritic neuron model (DNM) is an approximate logical neuron model based on dendrites, with branches of dendrites corresponding to three distributions in coordinates.The model is trained to assort data as needed by mimicking the mechanisms of transmitting information and biological nerves. Traditionally DNM models use error back propagation (BP) to optimise local minimum problems, but also degrade their performance. We now train it using an equilibrium optimizer based on physical phenomena inspired by control volume mass balance. Experimental results due to some real-world classification problems show that the mentioned algorithm can improve the accuracy of the DNM solution.
{"title":"Evolving Dendritic Neuron Model by Equilibrium Optimizer Algorithm","authors":"Chunzhi Hou, Jiarui Shi, Baohang Zhang","doi":"10.1109/PIC53636.2021.9687084","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687084","url":null,"abstract":"In recent years, the role of a single dendritic neural structures with non-linear localisation in computing has attracted a lot of attention from the industry. The dendritic neuron model (DNM) is an approximate logical neuron model based on dendrites, with branches of dendrites corresponding to three distributions in coordinates.The model is trained to assort data as needed by mimicking the mechanisms of transmitting information and biological nerves. Traditionally DNM models use error back propagation (BP) to optimise local minimum problems, but also degrade their performance. We now train it using an equilibrium optimizer based on physical phenomena inspired by control volume mass balance. Experimental results due to some real-world classification problems show that the mentioned algorithm can improve the accuracy of the DNM solution.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115856448","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687035
Kaiyu Dai, Yiyang Qiu, Rui Zhang
With the deepening integration of artificial intelligence, ICT in education is approaching to the stage of smart education, the main purpose of which is to realize learning personalization. This paper constructs an intelligent tutoring system to allow teacher establish the course knowledge model visually based on ontology. This system evaluates the learning situation of students using a test auto-generated by a global prediction accuracy optimization algorithm. The learning diagnosis module is implemented according to the learning situations of students and the structure analysis of knowledge graph based on node contribution. The resource recommendation module is implemented through the importance ranking of learning resources. The prototype system is constructed and the experiments are conducted. The results show that our approach can achieve personalized learning well in a certain range.
{"title":"The Construction of Learning Diagnosis and Resources Recommendation System Based on Knowledge Graph","authors":"Kaiyu Dai, Yiyang Qiu, Rui Zhang","doi":"10.1109/PIC53636.2021.9687035","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687035","url":null,"abstract":"With the deepening integration of artificial intelligence, ICT in education is approaching to the stage of smart education, the main purpose of which is to realize learning personalization. This paper constructs an intelligent tutoring system to allow teacher establish the course knowledge model visually based on ontology. This system evaluates the learning situation of students using a test auto-generated by a global prediction accuracy optimization algorithm. The learning diagnosis module is implemented according to the learning situations of students and the structure analysis of knowledge graph based on node contribution. The resource recommendation module is implemented through the importance ranking of learning resources. The prototype system is constructed and the experiments are conducted. The results show that our approach can achieve personalized learning well in a certain range.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"54 7","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"113970618","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}
Music emotion recognition (MER) has attracted much interest in the past decades for efficient music information organization and retrieval. Although deep learning has been applied to this field to avoid facing the complexity of feature engineering, the processing of original information within music pieces has become another challenge. In this paper, we propose a novel method named Frequency Embedded Regularization Network (FERN) for continuous MER to overcome this issue. Specifically, we apply regularized ResNet to automatically extract features through spectrograms with embedded frequency channels. The receptive fields in the deep architecture are adjusted by modifying the kernel size to maintain original information completely. Furthermore, Long Short-Term Memory (LSTM) is employed to learn the sequential relationship from the extracted contextual features. We conduct experiments on the benchmark dataset 1000 Songs. The experimental results show that our method is superior to most of the compared methods in terms of extracting salient features and catching the distribution of emotions within music pieces.
{"title":"Frequency Embedded Regularization Network for Continuous Music Emotion Recognition","authors":"Meixian Zhang, Yonghua Zhu, Ning Ge, Yunwen Zhu, Tianyu Feng, Wenjun Zhang","doi":"10.1109/PIC53636.2021.9687003","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687003","url":null,"abstract":"Music emotion recognition (MER) has attracted much interest in the past decades for efficient music information organization and retrieval. Although deep learning has been applied to this field to avoid facing the complexity of feature engineering, the processing of original information within music pieces has become another challenge. In this paper, we propose a novel method named Frequency Embedded Regularization Network (FERN) for continuous MER to overcome this issue. Specifically, we apply regularized ResNet to automatically extract features through spectrograms with embedded frequency channels. The receptive fields in the deep architecture are adjusted by modifying the kernel size to maintain original information completely. Furthermore, Long Short-Term Memory (LSTM) is employed to learn the sequential relationship from the extracted contextual features. We conduct experiments on the benchmark dataset 1000 Songs. The experimental results show that our method is superior to most of the compared methods in terms of extracting salient features and catching the distribution of emotions within music pieces.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133749109","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687034
Mulugeta Weldezgina Asres, G. Cummings, P. Parygin, A. Khukhunaishvili, M. Toms, A. Campbell, S. Cooper, D. Yu, J. Dittmann, C. Omlin
The ever-increasing detector complexity at CERN triggers a call for an increasing level of automation. Since the quality of collected physics data hinges on the quality of the detector components at the time of data-taking, the rapid identification and resolution of detector system anomalies will result in a better amount of high-quality particle data. Therefore, this study proposes CGVAE, a data-driven unsupervised anomaly detection using a deep learning model, for detector system monitoring from multivariate time series sensor data. The CGVAE model is composed of a variational autoencoder with convolutional and gated recurrent unit networks for fast localized feature extraction, long temporal characteristics capturing, and descriptive representation learning. Furthermore, to mitigate signal reconstruction overfitting on anomalous patterns, the CGVAE employs encoded latent feature- and reconstruction-based metrics for anomaly detection. Moreover, the model integrates feature attribution algorithms to explain the contribution of the input sensors to the detected anomalies. The experimental evaluation on large sensor data sets of the Hadron Calorimeter of the CMS experiment demonstrates the efficacy of the proposed model in capturing temporal anomalies.
{"title":"Unsupervised Deep Variational Model for Multivariate Sensor Anomaly Detection","authors":"Mulugeta Weldezgina Asres, G. Cummings, P. Parygin, A. Khukhunaishvili, M. Toms, A. Campbell, S. Cooper, D. Yu, J. Dittmann, C. Omlin","doi":"10.1109/PIC53636.2021.9687034","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687034","url":null,"abstract":"The ever-increasing detector complexity at CERN triggers a call for an increasing level of automation. Since the quality of collected physics data hinges on the quality of the detector components at the time of data-taking, the rapid identification and resolution of detector system anomalies will result in a better amount of high-quality particle data. Therefore, this study proposes CGVAE, a data-driven unsupervised anomaly detection using a deep learning model, for detector system monitoring from multivariate time series sensor data. The CGVAE model is composed of a variational autoencoder with convolutional and gated recurrent unit networks for fast localized feature extraction, long temporal characteristics capturing, and descriptive representation learning. Furthermore, to mitigate signal reconstruction overfitting on anomalous patterns, the CGVAE employs encoded latent feature- and reconstruction-based metrics for anomaly detection. Moreover, the model integrates feature attribution algorithms to explain the contribution of the input sensors to the detected anomalies. The experimental evaluation on large sensor data sets of the Hadron Calorimeter of the CMS experiment demonstrates the efficacy of the proposed model in capturing temporal anomalies.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"1995 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125554148","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687001
Francis Jonathan, Dong Yang, Glyn Gowing, Songjie Wei
Languages morphological context varies by community. The linguistic analysis became more complex due to grammatical variations, cultural, traditional, slang, misspellings, and language variance. Many studies in sentimental analysis have focused on natural language processing and peoples opinions. Text language processing takes time, requires lots of storage space, and a fast computer to work in distributed networks. Many developers choose Hadoop and Map Reduce to process Big Data. This study developed a methodology that employs Apache Spark as a text classification processing engine since it is faster in cluster computing systems. African libraries and packages for language lemmatization and stemming are still lacking. The proposed approach was utilized to detect offensive Swahili texts in social networks. Swahili is the third most widely spoken language in Africa. Four different machine learning techniques were tested as benchmarks, with the multinomial logistic model proving to be the most effective. The evaluation measures show that the proposed machine learning framework is versatile and suitable for usage in centralized and distributed systems.
{"title":"Machine Learning Framework for Detecting Offensive Swahili Messages in Social Networks with Apache Spark Implementation","authors":"Francis Jonathan, Dong Yang, Glyn Gowing, Songjie Wei","doi":"10.1109/PIC53636.2021.9687001","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687001","url":null,"abstract":"Languages morphological context varies by community. The linguistic analysis became more complex due to grammatical variations, cultural, traditional, slang, misspellings, and language variance. Many studies in sentimental analysis have focused on natural language processing and peoples opinions. Text language processing takes time, requires lots of storage space, and a fast computer to work in distributed networks. Many developers choose Hadoop and Map Reduce to process Big Data. This study developed a methodology that employs Apache Spark as a text classification processing engine since it is faster in cluster computing systems. African libraries and packages for language lemmatization and stemming are still lacking. The proposed approach was utilized to detect offensive Swahili texts in social networks. Swahili is the third most widely spoken language in Africa. Four different machine learning techniques were tested as benchmarks, with the multinomial logistic model proving to be the most effective. The evaluation measures show that the proposed machine learning framework is versatile and suitable for usage in centralized and distributed systems.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127628641","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}
Pub Date : 2021-12-17DOI: 10.1109/PIC53636.2021.9687015
Yufeng Ling, Jian Lu, Jian Dong, Tianjian Li, Zhiming Cai
Aiming at the problems of complex network structure, long training time and insufficient feature learning ability for deep learning, a lightweight network structure is designed. A kind of new activation function (namely rectifying linear unit) whose adaptive parameter is achieved by simplified training is proposed. The activation function is inserted into convolutional neural network to improve the feature learning ability by making each input signal has its own set of nonlinear transformation. Compared with traditional convolutional neural network, the number of network parameters is reduced by 51.61%, while the structure remains the ability of feature extraction before simplification. The proposed network structure can greatly reduce the network training time and improve the target recognition speed. The experiments on CIFAR-10 and CIFAR-100 datasets respectively show that the accuracies reach 95.26% and 76.54%, which are 1.67% and 3.76% higher than those of the traditional convolutional neural network.
{"title":"Wide Residual Lightweight Network Using Simplified Adaptive Parameter Rectifying Units","authors":"Yufeng Ling, Jian Lu, Jian Dong, Tianjian Li, Zhiming Cai","doi":"10.1109/PIC53636.2021.9687015","DOIUrl":"https://doi.org/10.1109/PIC53636.2021.9687015","url":null,"abstract":"Aiming at the problems of complex network structure, long training time and insufficient feature learning ability for deep learning, a lightweight network structure is designed. A kind of new activation function (namely rectifying linear unit) whose adaptive parameter is achieved by simplified training is proposed. The activation function is inserted into convolutional neural network to improve the feature learning ability by making each input signal has its own set of nonlinear transformation. Compared with traditional convolutional neural network, the number of network parameters is reduced by 51.61%, while the structure remains the ability of feature extraction before simplification. The proposed network structure can greatly reduce the network training time and improve the target recognition speed. The experiments on CIFAR-10 and CIFAR-100 datasets respectively show that the accuracies reach 95.26% and 76.54%, which are 1.67% and 3.76% higher than those of the traditional convolutional neural network.","PeriodicalId":297239,"journal":{"name":"2021 IEEE International Conference on Progress in Informatics and Computing (PIC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126663164","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：