Athlete body type recognition based on spectral co-clustering and CNN

Purpose This study aims to achieve an accurate body type classification of professional athletes. Design/methodology/approach In order to achieve body type classification of professional athletes, this article proposes a body type recognition method using a combination of joint spectral clustering, convolutional neural network and Taguchi test. Findings The results showed that the athletes' body types could be classified into four categories: small and compact (25.13%), tall and fit (26.67%), evenly proportioned and fit (24.62%) and limber (23.59%); the accuracy of the optimized convolutional neural network model was 99.43% and 97.14% in the training and test sets, respectively, and the loss rate was 2.49% and 5.12%, respectively. Originality/value The study is useful in facilitating research on the segmentation of professional athletes' body types and has practical value for the development of sportswear equipment. It also has some significance to the current research on body type classification and image recognition.