Pub Date : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126453
Akula Kumar Raja, Chidakash Sugandhi, Gorantla Nymish, Nama Sai Havish, Manazhy Rashmi
A disabled person’s life is always dependent on someone else who needs aid with mobility or any other task. Individuals with disabilities may face challenges when using computers. The most common way of interacting with computers is using a mouse and keyboard. It is difficult for people with physical disabilities to use them. Facial movements are one of the best possible actions by physically disabled individuals, By recognizing and responding to these movements, it is possible for them to operate the computer using only their facial expressions. Face recognition is a contemporary approach to interaction between humans and computers (HCI) i.e., The proposed system can easily control the computers by using face gesture recognition. It can be a viable replacement for traditional HCI tools in the future. This research outlines the techniques utilized in the design, implementation, and evaluation of the experiments conducted and presents the results obtained.
{"title":"Face Gesture Based Virtual Mouse Using Mediapipe","authors":"Akula Kumar Raja, Chidakash Sugandhi, Gorantla Nymish, Nama Sai Havish, Manazhy Rashmi","doi":"10.1109/I2CT57861.2023.10126453","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126453","url":null,"abstract":"A disabled person’s life is always dependent on someone else who needs aid with mobility or any other task. Individuals with disabilities may face challenges when using computers. The most common way of interacting with computers is using a mouse and keyboard. It is difficult for people with physical disabilities to use them. Facial movements are one of the best possible actions by physically disabled individuals, By recognizing and responding to these movements, it is possible for them to operate the computer using only their facial expressions. Face recognition is a contemporary approach to interaction between humans and computers (HCI) i.e., The proposed system can easily control the computers by using face gesture recognition. It can be a viable replacement for traditional HCI tools in the future. This research outlines the techniques utilized in the design, implementation, and evaluation of the experiments conducted and presents the results obtained.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117249915","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126449
Niloofar Abed, Ramu Murugan
The popularity and prevalence of devices equipped with object detection technology and controllable via the Internet of Things (IoT) have increased, especially in the post-Corona era. The development of neural networks and artificial intelligence by combining them with IoT systems has achieved acceptable satisfaction among users in adverse conditions by reducing the need for manpower and increasing productivity. Therefore, the scope of using such mechanisms has expanded in most fields, from self-driving vehicles to agricultural crops. Beginners will be confronted with a massive amount of complex information as a result of the design and application of such technologies in interdisciplinary fields. Due to the popularity of using the You Only Look Once (YOLO) object detection algorithm, this article provided a guideline as a traffic light subject classification and, offers suggested solutions and exclusive approches to increase the accuracy of object detection in real-time projects with a practical application attitude for the enthusiasts and developers particularly in object detection scenarios by employing YOLO.
配备目标检测技术并通过物联网(IoT)控制的设备的普及程度和流行程度有所增加，特别是在后冠状病毒时代。神经网络和人工智能的发展与物联网系统相结合，减少了对人力的需求，提高了生产率，在不利的条件下，用户获得了可接受的满意度。因此，从自动驾驶汽车到农作物，这种机制的使用范围在大多数领域都得到了扩展。由于这些技术在跨学科领域的设计和应用，初学者将面临大量复杂的信息。由于YOLO (You Only Look Once)目标检测算法的普及，本文作为红绿灯主题分类的指南，以实际应用的态度，为使用YOLO进行目标检测的爱好者和开发者提供了提高实时项目中目标检测精度的建议解决方案和独家途径。
{"title":"Strategies for Improving Object Detection in Real-Time Projects that use Deep Learning Technology","authors":"Niloofar Abed, Ramu Murugan","doi":"10.1109/I2CT57861.2023.10126449","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126449","url":null,"abstract":"The popularity and prevalence of devices equipped with object detection technology and controllable via the Internet of Things (IoT) have increased, especially in the post-Corona era. The development of neural networks and artificial intelligence by combining them with IoT systems has achieved acceptable satisfaction among users in adverse conditions by reducing the need for manpower and increasing productivity. Therefore, the scope of using such mechanisms has expanded in most fields, from self-driving vehicles to agricultural crops. Beginners will be confronted with a massive amount of complex information as a result of the design and application of such technologies in interdisciplinary fields. Due to the popularity of using the You Only Look Once (YOLO) object detection algorithm, this article provided a guideline as a traffic light subject classification and, offers suggested solutions and exclusive approches to increase the accuracy of object detection in real-time projects with a practical application attitude for the enthusiasts and developers particularly in object detection scenarios by employing YOLO.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115886182","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126335
V. Khatavkar, Snehal Andhale, Panchshila Pillewar, Utkarsh Alset
The wind turbine requires a robust and time-responsive system to control the pitch–angle (Pit–Ang) of the mechanical actuator. If the response of speed is very efficient then, the controller can act according to the prescribed logic and its mechanical mechanism can work faster with its response time. In this paper, real-time Data from IMD (Indian Meteorological Department) is used for the relative study of the model of wind turbine created in MATLAB / Simulink® environment using Fuzzy Logic Toolbox™. The principle of wind turbine used here is to supply a controlled input to the system and after synthesis, these rules in the form of signal are transferred to the plant which has a drive train and pitch actuator. The output responses of the proposed controller are compared amongst proportional– integral–derivative (PID), fuzzy, and adaptive fuzzy–PID Controllers. The simulation results seen between the adaptive fuzzy– based PID controller surpasses the expected results by Tr = 95.454%, Ts = 61.409% and negligible overshoot as compared to open–looped and conventional responsive controller.
{"title":"Relative Study of Intelligent Control Techniques to Maintain Variable Pitch-Angle of the Wind Turbine","authors":"V. Khatavkar, Snehal Andhale, Panchshila Pillewar, Utkarsh Alset","doi":"10.1109/I2CT57861.2023.10126335","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126335","url":null,"abstract":"The wind turbine requires a robust and time-responsive system to control the pitch–angle (Pit–Ang) of the mechanical actuator. If the response of speed is very efficient then, the controller can act according to the prescribed logic and its mechanical mechanism can work faster with its response time. In this paper, real-time Data from IMD (Indian Meteorological Department) is used for the relative study of the model of wind turbine created in MATLAB / Simulink® environment using Fuzzy Logic Toolbox™. The principle of wind turbine used here is to supply a controlled input to the system and after synthesis, these rules in the form of signal are transferred to the plant which has a drive train and pitch actuator. The output responses of the proposed controller are compared amongst proportional– integral–derivative (PID), fuzzy, and adaptive fuzzy–PID Controllers. The simulation results seen between the adaptive fuzzy– based PID controller surpasses the expected results by Tr = 95.454%, Ts = 61.409% and negligible overshoot as compared to open–looped and conventional responsive controller.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"28 9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115634459","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126428
Hridya V Ramesh, Rahul Krishnan Pathinarupothi
Globally the rate of heart disease has increased drastically due to unhealthy eating habits and reduced physical activities. It has become one of the significant causes of death worldwide. As per the reports of the world health organization(WHO), 31% of all deaths worldwide are caused by cardiovascular diseases. This demands the development of a system capable of early detection of cardiovascular diseases at an affordable cost. With this as the objective, multiple machine learning algorithms have been selected to evaluate their performance in the early detection of cardiovascular diseases. This work utilizes available data sets of an individual’s vital parameters, demographic data, and exercise parameters for predicting cardiovascular diseases. An extensive evaluation is performed to identify the best-suited supervised machine learning classifier that could predict cardiovascular diseases using the available datasets. This research work details the nine different classification algorithms utilized for this analysis. For each algorithm, the F1-score, precision, recall, accuracy, and Area Under the Receiver Operating Characteristics (AUROC) values for each model have been determined and compared with the rest of the algorithms. The results show that random forest and gradient boosting models outperform others and demonstrate an F1-Score of 0.88 and an AUROC value of 0.92, respectively. This showcases that doctors could utilize this technique for the early identification of cardiovascular diseases. This will provide the opportunity to offer adequate medical treatments early, thus saving lives.
在全球范围内，由于不健康的饮食习惯和体育活动的减少，心脏病的发病率急剧上升。它已成为世界范围内死亡的主要原因之一。根据世界卫生组织(WHO)的报告，全世界31%的死亡是由心血管疾病引起的。这就要求开发一种能够以负担得起的成本及早发现心血管疾病的系统。以此为目标，我们选择了多种机器学习算法来评估它们在心血管疾病早期检测中的表现。这项工作利用个人重要参数、人口统计数据和运动参数的可用数据集来预测心血管疾病。进行了广泛的评估，以确定最适合的监督机器学习分类器，该分类器可以使用可用的数据集预测心血管疾病。这项研究工作详细介绍了用于此分析的九种不同的分类算法。对于每种算法，确定了每种模型的f1评分、精度、召回率、准确度和接收者操作特征下面积(Area Under the Receiver Operating Characteristics, AUROC)值，并与其他算法进行了比较。结果表明，随机森林模型和梯度增强模型的F1-Score为0.88,AUROC值为0.92，优于其他模型。这表明医生可以利用这项技术来早期识别心血管疾病。这将提供机会及早提供适当的医疗，从而挽救生命。
{"title":"Performance Analysis of Machine Learning Algorithms to Predict Cardiovascular Disease","authors":"Hridya V Ramesh, Rahul Krishnan Pathinarupothi","doi":"10.1109/I2CT57861.2023.10126428","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126428","url":null,"abstract":"Globally the rate of heart disease has increased drastically due to unhealthy eating habits and reduced physical activities. It has become one of the significant causes of death worldwide. As per the reports of the world health organization(WHO), 31% of all deaths worldwide are caused by cardiovascular diseases. This demands the development of a system capable of early detection of cardiovascular diseases at an affordable cost. With this as the objective, multiple machine learning algorithms have been selected to evaluate their performance in the early detection of cardiovascular diseases. This work utilizes available data sets of an individual’s vital parameters, demographic data, and exercise parameters for predicting cardiovascular diseases. An extensive evaluation is performed to identify the best-suited supervised machine learning classifier that could predict cardiovascular diseases using the available datasets. This research work details the nine different classification algorithms utilized for this analysis. For each algorithm, the F1-score, precision, recall, accuracy, and Area Under the Receiver Operating Characteristics (AUROC) values for each model have been determined and compared with the rest of the algorithms. The results show that random forest and gradient boosting models outperform others and demonstrate an F1-Score of 0.88 and an AUROC value of 0.92, respectively. This showcases that doctors could utilize this technique for the early identification of cardiovascular diseases. This will provide the opportunity to offer adequate medical treatments early, thus saving lives.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"114 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116332368","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126370
Shubham Sah, S. Prakash, S. Meena
Everyone recalls the destruction brought on by the Australian forest fires in 2019. Even though there isn’t much we can do to battle forest fires on our own, we can always rely on technology. By this we are trying to predict the accuracy of these models on forest fire data set. We are trying to detect forest fire in dense forest; our data set is very diverse and consist of images having forest fires, smokes, non-smoke and fire images. We have found out that Sensor detection and real-time geological data analysis are two methods for detecting forest fires. However, using image classification, for which Deep learning is the most efficient solution, is one of the best methods for detecting fire. In addition, these algorithms can be integrated with drones using deep learning techniques so that images can be taken frequently from the sky with ease, smoke can be detected in dense forests, and the authorities can be notified to take immediate action. The convolutional neural network algorithm for fire detection was the sole focus of our paper. The value of various epochs is used to evaluate these results.
{"title":"Forest Fire Detection using Convolutional Neural Network Model","authors":"Shubham Sah, S. Prakash, S. Meena","doi":"10.1109/I2CT57861.2023.10126370","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126370","url":null,"abstract":"Everyone recalls the destruction brought on by the Australian forest fires in 2019. Even though there isn’t much we can do to battle forest fires on our own, we can always rely on technology. By this we are trying to predict the accuracy of these models on forest fire data set. We are trying to detect forest fire in dense forest; our data set is very diverse and consist of images having forest fires, smokes, non-smoke and fire images. We have found out that Sensor detection and real-time geological data analysis are two methods for detecting forest fires. However, using image classification, for which Deep learning is the most efficient solution, is one of the best methods for detecting fire. In addition, these algorithms can be integrated with drones using deep learning techniques so that images can be taken frequently from the sky with ease, smoke can be detected in dense forests, and the authorities can be notified to take immediate action. The convolutional neural network algorithm for fire detection was the sole focus of our paper. The value of various epochs is used to evaluate these results.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115472819","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126406
Prachi V. Karlekar, Swapna Choudhary, Atul Deshmukh, Harish Banote
Robotic arm movements are highly dependent on design and deployment of sensors & actuation devices & their duty cycles. Optimizing current-level duty cycles for these devices can reduce the power consumption, and maximize the efficiency of control for different device operations. Existing duty cycle control models for robotic arms are highly complex, or have lower efficiency levels. To overcome these issues, this text proposes design of an efficient bioinspired model for optimizing robotic arm movements via ensemble learning operations. The arm is built using Arduino controller along with stepper motors, which assist in controlled movements for different arm operations. The proposed model uses Mayfly Optimization (MO) in order to identify duty cycles of different arm components for different movement types. The MO Model uses delay, energy and jitter parameters in order to estimate a fitness function that is optimized in order to identify arm movement sets. These movement sets are classified into performance-aware movements via a combination of Naïve Bayes (NB), k Nearest Neighbours (kNN), Support Vector Machine (SVM), Logistic Regression (LR), and Multilayer Perceptron (MLP) classifiers. Due to which the model is able to reduce the delay needed for control the arms by 8.3%, reduce the energy needed for control operations by 2.9%, and reduce the control jitter by 4.5% under real-time scenarios.
{"title":"Design of an Efficient Bioinspired Model for Optimizing Robotic Arm Movements via Ensemble Learning Operations","authors":"Prachi V. Karlekar, Swapna Choudhary, Atul Deshmukh, Harish Banote","doi":"10.1109/I2CT57861.2023.10126406","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126406","url":null,"abstract":"Robotic arm movements are highly dependent on design and deployment of sensors & actuation devices & their duty cycles. Optimizing current-level duty cycles for these devices can reduce the power consumption, and maximize the efficiency of control for different device operations. Existing duty cycle control models for robotic arms are highly complex, or have lower efficiency levels. To overcome these issues, this text proposes design of an efficient bioinspired model for optimizing robotic arm movements via ensemble learning operations. The arm is built using Arduino controller along with stepper motors, which assist in controlled movements for different arm operations. The proposed model uses Mayfly Optimization (MO) in order to identify duty cycles of different arm components for different movement types. The MO Model uses delay, energy and jitter parameters in order to estimate a fitness function that is optimized in order to identify arm movement sets. These movement sets are classified into performance-aware movements via a combination of Naïve Bayes (NB), k Nearest Neighbours (kNN), Support Vector Machine (SVM), Logistic Regression (LR), and Multilayer Perceptron (MLP) classifiers. Due to which the model is able to reduce the delay needed for control the arms by 8.3%, reduce the energy needed for control operations by 2.9%, and reduce the control jitter by 4.5% under real-time scenarios.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116748604","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126460
Rupali J Dhabarde, D. V. Kodawade, Sheetal Zalte
A technique for recognizing a person from his photograph is facial recognition. Due to its extensive range of applications in several fields, it has drawn the attention of numerous researchers in the field of computer vision in recent years (Cyber security, crime cases, and biometrics). This technology's operation is based on the extraction of features from an input picture using methods like PCA, ICA, LDA etc. After comparing them with others from another image to verify or assert an individual's identification. Via this work, we applied amalgamation of CNN and SVM techniques to two face datasets that will be split into two groups in a machine learning-based methodology. We assessed different machine learning-based lie detectors using our amassed dataset. Our findings demonstrate that combined CNN with SVM task achieved accuracy up to 58%.
{"title":"Hybrid Machine Learning Model for Lie-Detection","authors":"Rupali J Dhabarde, D. V. Kodawade, Sheetal Zalte","doi":"10.1109/I2CT57861.2023.10126460","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126460","url":null,"abstract":"A technique for recognizing a person from his photograph is facial recognition. Due to its extensive range of applications in several fields, it has drawn the attention of numerous researchers in the field of computer vision in recent years (Cyber security, crime cases, and biometrics). This technology's operation is based on the extraction of features from an input picture using methods like PCA, ICA, LDA etc. After comparing them with others from another image to verify or assert an individual's identification. Via this work, we applied amalgamation of CNN and SVM techniques to two face datasets that will be split into two groups in a machine learning-based methodology. We assessed different machine learning-based lie detectors using our amassed dataset. Our findings demonstrate that combined CNN with SVM task achieved accuracy up to 58%.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115580918","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126397
D. Banerjee, V. Kukreja, S. Hariharan, Vishal Jain
The mango leaf powdery mildew disease poses a serious threat to mango production society globally by significantly lowering yield and quality. For timely intervention and efficient management, early disease detection and classification are important. In this research and education area, a novel hybrid approach utilizes Convolutional Neural Networks (CNN) and Support Vector Machines to identify the mango leaf powdery mildew disease based on four severity levels (SVM). Three phases make up the proposed approach: data structure, CNN-selected attributes, and SVM classification. We collect and preprocess images of mango leaves during the data organization step, and in the CNN - attributes selection phase, we apply a CNN model for feature extraction and selection. For the mango leaf powdery mildew dataset, we improve the CNN model to find the most relevant features for the classification task. The SVM - classification step includes training an SVM model on the obtained features and refining the hyperparameters via k-fold cross-validation. The proposed CNN and SVM hybrid multi-classification model for mango leaf powdery mildew disease achieved an overall accuracy of 89.29%. A dataset of 2559 images with 4 severity levels was utilized. The model works well overall, as a macro-average F1-score of 90.10, the weighted average F1-score's minimal value of 53.85%. The model is less successful in predicting instances for classes with smaller support proportions, as shown by the micro-average F1-score, which is 89.29% and is lower overall than the macro-average F1-score.
{"title":"Enhancing Mango Fruit Disease Severity Assessment with CNN and SVM-Based Classification","authors":"D. Banerjee, V. Kukreja, S. Hariharan, Vishal Jain","doi":"10.1109/I2CT57861.2023.10126397","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126397","url":null,"abstract":"The mango leaf powdery mildew disease poses a serious threat to mango production society globally by significantly lowering yield and quality. For timely intervention and efficient management, early disease detection and classification are important. In this research and education area, a novel hybrid approach utilizes Convolutional Neural Networks (CNN) and Support Vector Machines to identify the mango leaf powdery mildew disease based on four severity levels (SVM). Three phases make up the proposed approach: data structure, CNN-selected attributes, and SVM classification. We collect and preprocess images of mango leaves during the data organization step, and in the CNN - attributes selection phase, we apply a CNN model for feature extraction and selection. For the mango leaf powdery mildew dataset, we improve the CNN model to find the most relevant features for the classification task. The SVM - classification step includes training an SVM model on the obtained features and refining the hyperparameters via k-fold cross-validation. The proposed CNN and SVM hybrid multi-classification model for mango leaf powdery mildew disease achieved an overall accuracy of 89.29%. A dataset of 2559 images with 4 severity levels was utilized. The model works well overall, as a macro-average F1-score of 90.10, the weighted average F1-score's minimal value of 53.85%. The model is less successful in predicting instances for classes with smaller support proportions, as shown by the micro-average F1-score, which is 89.29% and is lower overall than the macro-average F1-score.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"164 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123135445","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126193
Raghavendra A Hallyal, S. C, P. Desai, M. M
Information retrieval from the CAPTCHA is a crucial part, this CAPTCHA always contains some unwanted information along with required information, so attention technique comes in handy to select such useful information discarding the unwanted part. The attention concept has become a very important part in the field of deep learning which uses Natural Language Processing(NLP) and Computer Vision(CV). Attention mechanism is rigorously used in OCR based applications which requires generating of selected information rather than every information available. Our work includes implementation of general, global and local Attention mechanisms used with two different models which were transfer learning model and the parameter search model. As OCR with attention technique is computationally costly it is required to optimize the entire process so we suggest optimized retrieval of information from CAPTCHA using parameter search algorithm. This retrieval includes using weights that reduced the training time from 4.03 minutes to 3.33 minutes and the number of training images which were used for training were reduced than before. We obtained the highest accuracy of 87.34% for general attention with parameter search model and local attention model with parameter search model proved to have less computation and less training time than the general attention with parameter search model.
{"title":"Optimized Recognition Of CAPTCHA Through Attention Models","authors":"Raghavendra A Hallyal, S. C, P. Desai, M. M","doi":"10.1109/I2CT57861.2023.10126193","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126193","url":null,"abstract":"Information retrieval from the CAPTCHA is a crucial part, this CAPTCHA always contains some unwanted information along with required information, so attention technique comes in handy to select such useful information discarding the unwanted part. The attention concept has become a very important part in the field of deep learning which uses Natural Language Processing(NLP) and Computer Vision(CV). Attention mechanism is rigorously used in OCR based applications which requires generating of selected information rather than every information available. Our work includes implementation of general, global and local Attention mechanisms used with two different models which were transfer learning model and the parameter search model. As OCR with attention technique is computationally costly it is required to optimize the entire process so we suggest optimized retrieval of information from CAPTCHA using parameter search algorithm. This retrieval includes using weights that reduced the training time from 4.03 minutes to 3.33 minutes and the number of training images which were used for training were reduced than before. We obtained the highest accuracy of 87.34% for general attention with parameter search model and local attention model with parameter search model proved to have less computation and less training time than the general attention with parameter search model.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123330931","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 : 2023-04-07DOI: 10.1109/I2CT57861.2023.10126396
Meenu Mathew, J. Prakash
Textual emotion detection is a classification problem that assigns different emotions to a given text input. It reveals the writer’s mental state. Its diversity and uncertainty make it a challenging task. The existing methods in machine learning can be used for emotion detection; however, it fails in processing very long passages. In this work, we employ weighted pooling pretrained RoBERTa model for effective textual emotion detection. To perform experiments, we use two data sets, ISEAR and tweets, with 7516 and 21048 records, respectively. Precision, recall, F1-score, and classification accuracy are used to assess the models. Experimental results indicate that the weighted pooling RoBERTa model outperforms the deep learning models on both datasets with significant improvement in accuracy.
{"title":"Weighted Pooling RoBERTa for Effective Text Emotion Detection","authors":"Meenu Mathew, J. Prakash","doi":"10.1109/I2CT57861.2023.10126396","DOIUrl":"https://doi.org/10.1109/I2CT57861.2023.10126396","url":null,"abstract":"Textual emotion detection is a classification problem that assigns different emotions to a given text input. It reveals the writer’s mental state. Its diversity and uncertainty make it a challenging task. The existing methods in machine learning can be used for emotion detection; however, it fails in processing very long passages. In this work, we employ weighted pooling pretrained RoBERTa model for effective textual emotion detection. To perform experiments, we use two data sets, ISEAR and tweets, with 7516 and 21048 records, respectively. Precision, recall, F1-score, and classification accuracy are used to assess the models. Experimental results indicate that the weighted pooling RoBERTa model outperforms the deep learning models on both datasets with significant improvement in accuracy.","PeriodicalId":150346,"journal":{"name":"2023 IEEE 8th International Conference for Convergence in Technology (I2CT)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123853003","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：