A DATASET OF 117 INDIVIDUALS WITH POOR POSTURE AND 80 INDIVIDUALS WITH NORMAL POSTURE

A poor posture is a common health issue for adolescents during their growth and development. A prolonged poor posture can lead to musculoskeletal pain and disorders, and may even affect adolescents' growth and development. However, it is time-consuming and subjective to assess the poor posture in adolescents. Thus it is crucial to obtain an accurate and rapid evaluation method for poor posture. This paper proposes a method for recognizing poor posture gait in adolescents based on multi-convolutional neural networks and convolutional block attention modules. In this method, the foot pressure gait is firstly segmented and one-dimensional foot pressure data is transformed into two-dimensional grayscale images (i.e., pressure data transform images, PDTI), and then the features are extracted by using two convolutional neural networks of different scales with a convolutional block attention module (CBAM) to identify the abnormal gaits. With 197 adolescent volunteers from Hunan Provincial People's Hospital (117 with poor posture, 80 with normal posture), their gait recognition is performed based on this proposed PDTI-CNNs-CBAM. Experimental results show that when the normal gait thresholds respectively are 60%, 70%, 80%, and 90%, the model's recognition accuracy is 100%, 98.1%, 94.6%, and 84.3%. In addition, comparing with CNN methods, CNN-CBAM methods, and traditional feature-based methods (such as SVM, BP, and RBFNN), the experimental results demonstrate that the accuracy of this proposedPDTI-CNNs-CBAM model is the highest, which confirms its effectiveness.