Pub Date : 2023-06-29DOI: 10.1109/SCSE59836.2023.10214999
Nipuna Munasinghe, S. Jayalal, T. Wijayasiriwardhane
Hearing-impaired and speech-impaired people communicate not only with themselves but also with ordinary people using visual languages. Sri Lankan Sign Language (SSL) is the standard visual language used in Sri Lanka. Like other sign languages, the SSL relies on a distinct combination of hand gestures, body movements, and facial expressions for communication. As a result, SSL is more challenging for individuals without knowledge of SSL to understand. On the other hand, the steep learning curve associated with SSL makes it even more difficult to acquire. Thus, the interpretation of SSL has become a need. However, Sri Lanka is suffering from a severe dearth in the availability of SSL interpreters. This justifies the need to use either vision-based or sensor-based technological approaches to help the interpretation of SSL. However, vision-based approaches are susceptible to conditions such as skin tone, background color, ambient light intensity, and real-time constraints, whilst the sensor-based solutions are generally better in gesture recognition. Further, there is no attempt has been made on developing a cost-effective, portable, and real-time solution to accurately interpret the hand gestures of SSL. In this paper, we, therefore, present a novel, wearable, sensor-based, real-time gesture recognition glove, and a machine-learning Long Short-Term Memory (LSTM) model to recognize the hand and finger positions in three-dimensional space for classification and interpretation of SSL. The proposed approach has achieved 320ms of lowest inference time while showing a promising result of 83% for categorical accuracy. Our aim is to help the interpretation of SSL with an affordable, portable as well as a real-time solution.
{"title":"Interpretation of Sri Lankan Sign Language: A Wearable Sensor-based Approach","authors":"Nipuna Munasinghe, S. Jayalal, T. Wijayasiriwardhane","doi":"10.1109/SCSE59836.2023.10214999","DOIUrl":"https://doi.org/10.1109/SCSE59836.2023.10214999","url":null,"abstract":"Hearing-impaired and speech-impaired people communicate not only with themselves but also with ordinary people using visual languages. Sri Lankan Sign Language (SSL) is the standard visual language used in Sri Lanka. Like other sign languages, the SSL relies on a distinct combination of hand gestures, body movements, and facial expressions for communication. As a result, SSL is more challenging for individuals without knowledge of SSL to understand. On the other hand, the steep learning curve associated with SSL makes it even more difficult to acquire. Thus, the interpretation of SSL has become a need. However, Sri Lanka is suffering from a severe dearth in the availability of SSL interpreters. This justifies the need to use either vision-based or sensor-based technological approaches to help the interpretation of SSL. However, vision-based approaches are susceptible to conditions such as skin tone, background color, ambient light intensity, and real-time constraints, whilst the sensor-based solutions are generally better in gesture recognition. Further, there is no attempt has been made on developing a cost-effective, portable, and real-time solution to accurately interpret the hand gestures of SSL. In this paper, we, therefore, present a novel, wearable, sensor-based, real-time gesture recognition glove, and a machine-learning Long Short-Term Memory (LSTM) model to recognize the hand and finger positions in three-dimensional space for classification and interpretation of SSL. The proposed approach has achieved 320ms of lowest inference time while showing a promising result of 83% for categorical accuracy. Our aim is to help the interpretation of SSL with an affordable, portable as well as a real-time solution.","PeriodicalId":429228,"journal":{"name":"2023 International Research Conference on Smart Computing and Systems Engineering (SCSE)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128774574","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 : 2023-06-29DOI: 10.1109/SCSE59836.2023.10215041
J. Jayakody, Vgtn Vidanagama, Indika Perera, Hmlk Herath
feature selection technique is used in text classification pipeline to reduce the number of redundant or irrelevant features. Moreover, feature selection algorithms help to decrease the overfitting, reduce training time and improve the accuracy of the build models. Similarly, feature reduction techniques based on frequencies support eliminating unwanted features. Most of the existing work related to feature selection was based on general text and the behavior of feature selection was not evaluated properly with short text type dataset. Therefore this research was conducted to investigate how performance varied with selected features from feature selection algorithms with short text type datasets. Three publicly available datasets were selected for the experiment. Chi square, info gain and f measure were examined as those algorithms were identified as the best algorithms to select features for text classification. Moreover, we examined the impact of those algorithms when selecting different types of features such as 1gram and 2-gram. Finally, we look at the impact of frequency based feature reduction techniques with the selected dataset. Our results showed that info gain algorithm outperform other two algorithms. Moreover, selection of best 20% feature set with info gain algorithm provide the same performance level as with the entire feature set. Further we observed the higher number of dimensions was due to bigrams and the impact of n grams towards feature selection algorithms. Moreover, it is worth noting that removing the features which occur twice in a document would be ideal before moving to apply feature selection techniques with different algorithms.
{"title":"Impact of Feature Selection Towards Short Text Classification","authors":"J. Jayakody, Vgtn Vidanagama, Indika Perera, Hmlk Herath","doi":"10.1109/SCSE59836.2023.10215041","DOIUrl":"https://doi.org/10.1109/SCSE59836.2023.10215041","url":null,"abstract":"feature selection technique is used in text classification pipeline to reduce the number of redundant or irrelevant features. Moreover, feature selection algorithms help to decrease the overfitting, reduce training time and improve the accuracy of the build models. Similarly, feature reduction techniques based on frequencies support eliminating unwanted features. Most of the existing work related to feature selection was based on general text and the behavior of feature selection was not evaluated properly with short text type dataset. Therefore this research was conducted to investigate how performance varied with selected features from feature selection algorithms with short text type datasets. Three publicly available datasets were selected for the experiment. Chi square, info gain and f measure were examined as those algorithms were identified as the best algorithms to select features for text classification. Moreover, we examined the impact of those algorithms when selecting different types of features such as 1gram and 2-gram. Finally, we look at the impact of frequency based feature reduction techniques with the selected dataset. Our results showed that info gain algorithm outperform other two algorithms. Moreover, selection of best 20% feature set with info gain algorithm provide the same performance level as with the entire feature set. Further we observed the higher number of dimensions was due to bigrams and the impact of n grams towards feature selection algorithms. Moreover, it is worth noting that removing the features which occur twice in a document would be ideal before moving to apply feature selection techniques with different algorithms.","PeriodicalId":429228,"journal":{"name":"2023 International Research Conference on Smart Computing and Systems Engineering (SCSE)","volume":"93 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125327895","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：