Article: Application of convolutional neural network in normative detection and error correction of college students' physical exercise actions Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.10 pp.72 - 87 Abstract: To address the issues of experience-dependent movement standard evaluation and delayed error correction feedback in college PE, a bond product neural network-based method is developed for detecting and correcting students' exercise movements. Focusing on squats, push-ups, and lunges, the study builds a skeleton sequence dataset, and adopts a model combining convolutional feature extraction and multi-distribution attention weights for joint movement classification, standard scoring, and deviation correction. Experimental results show that the model's action classification accuracy on the test set is 92.7%-96.8%, with an average absolute error of 2.95 and a reasoning delay of ~11 ms. Normative scores are 77.6-84.1; 28.1% and 25.4% of corrections target insufficient squat depth and hip sinking. The results demonstrate that this method meets the accuracy and real-time demands of college PE classrooms, providing quantitative support for teaching evaluation and training intervention. Inderscience Publishers - linking academia, business and industry through research

Title: Application of convolutional neural network in normative detection and error correction of college students' physical exercise actions

Authors: Huan Chen

Addresses: School of Physical Education, Ankang University, Ankang, 725000, China

Abstract: To address the issues of experience-dependent movement standard evaluation and delayed error correction feedback in college PE, a bond product neural network-based method is developed for detecting and correcting students' exercise movements. Focusing on squats, push-ups, and lunges, the study builds a skeleton sequence dataset, and adopts a model combining convolutional feature extraction and multi-distribution attention weights for joint movement classification, standard scoring, and deviation correction. Experimental results show that the model's action classification accuracy on the test set is 92.7%-96.8%, with an average absolute error of 2.95 and a reasoning delay of ~11 ms. Normative scores are 77.6-84.1; 28.1% and 25.4% of corrections target insufficient squat depth and hip sinking. The results demonstrate that this method meets the accuracy and real-time demands of college PE classrooms, providing quantitative support for teaching evaluation and training intervention.

Keywords: coupon product neural network; college students; physical education in colleges and universities; normative score of action; error correction prompt.

DOI: 10.1504/IJRIS.2026.152546

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.10, pp.72 - 87

Received: 19 Dec 2025
Accepted: 21 Jan 2026
Published online: 26 Mar 2026 *

Title: Application of convolutional neural network in normative detection and error correction of college students' physical exercise actions

Keep up-to-date