Scholarlink

<p>Yoga is not only a physical practice, it is also a holistic way to living healthy life. In recent years, practicing yoga has been essential to maintaining physical and mental wellness. Although practicing yoga under the guidance is ideal for maximizing benefits but fast-paced lifestyle and the high cost of personal trainers can make this problematic especially for middle-class families. When individuals choose to practice yoga on their own, they often face the challenge of maintaining correct posture. Without proper guidance incorrect posture can lead long term injuries. To overcome this challenges, the AI- based guidance system can help. This paper introduces a hybrid LGBM-MLP based classification (YogLiMLP) model for Yoga Poses identification. In the pre-processing stage reshaping method is applied on image dataset to better analysis processing. Shearing and zooming augmentation techniques is employed for enhancing data size. Here, for the skeletonization process MoveNet pose estimation method is used. With the help of MoveNet method a kinematic model of the human body is generated by establishing connections between 17 keypoints, providing a visual representation of the skeletal structure. This method will help to recognize correct yoga poses and determine how well yoga postures are performed within computer vision. The classification stage employs ensemble based learning model using two classifiers (The Keras-based Multilayer Perceptron and Light Gradient Boosting Machine) to identify accurate yoga pose for individuals. The Keras-based Multilayer Perceptron (KMLP), used to learn complex patterns that cannot be represented by linear models with Light Gradient Boosting Machine (LGBM) boosting capabilities offers a robust, scalable solution for yoga pose classification tasks. The proposed model handles large datasets, mitigates overfitting, reduces computational overhead, cost and speeds up the training process. The performance of proposed architecture is well on a publicly accessible yoga pose dataset using four standard metrics. YogLiMLP model achieves the high classification accuracy (99.45%) compared to other State-Of-The-Art (SOTA) methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

YogLiMLP: Hybrid LGBM-MLP based classification model for yoga poses

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Jyoti Jangade,
  • Kanojia Sindhuben Babulal

摘要

Yoga is not only a physical practice, it is also a holistic way to living healthy life. In recent years, practicing yoga has been essential to maintaining physical and mental wellness. Although practicing yoga under the guidance is ideal for maximizing benefits but fast-paced lifestyle and the high cost of personal trainers can make this problematic especially for middle-class families. When individuals choose to practice yoga on their own, they often face the challenge of maintaining correct posture. Without proper guidance incorrect posture can lead long term injuries. To overcome this challenges, the AI- based guidance system can help. This paper introduces a hybrid LGBM-MLP based classification (YogLiMLP) model for Yoga Poses identification. In the pre-processing stage reshaping method is applied on image dataset to better analysis processing. Shearing and zooming augmentation techniques is employed for enhancing data size. Here, for the skeletonization process MoveNet pose estimation method is used. With the help of MoveNet method a kinematic model of the human body is generated by establishing connections between 17 keypoints, providing a visual representation of the skeletal structure. This method will help to recognize correct yoga poses and determine how well yoga postures are performed within computer vision. The classification stage employs ensemble based learning model using two classifiers (The Keras-based Multilayer Perceptron and Light Gradient Boosting Machine) to identify accurate yoga pose for individuals. The Keras-based Multilayer Perceptron (KMLP), used to learn complex patterns that cannot be represented by linear models with Light Gradient Boosting Machine (LGBM) boosting capabilities offers a robust, scalable solution for yoga pose classification tasks. The proposed model handles large datasets, mitigates overfitting, reduces computational overhead, cost and speeds up the training process. The performance of proposed architecture is well on a publicly accessible yoga pose dataset using four standard metrics. YogLiMLP model achieves the high classification accuracy (99.45%) compared to other State-Of-The-Art (SOTA) methods.