The sport data

1.Dataset overview

This dataset is designed to support the HAR task of this study. Covered by (a) walking, (b) running, (c) going upstairs, (d) going downstairs, (e) high leg lifting, (f) skipping rope, and (g) rhombic extension Seven types of human movement data.

The files D.xlsx, E.xlsx, H.xlsx, R.xlsx, S.xlsx, U.xlsx, and W.xlsx are one-dimensional time series data of seven sports, with a data length of 400, and the number of data categories of 7.

The data contained in folders D, E, H, R, S, U, and W are respectively two-dimensional image data converted from one-dimensional time series in files D.lsx, e.lsx, h.lsx, r.lsx, s.lsx, u.lsx, and w.lsx using the GAF algorithm. The resolution of the images is 224×224. The number of color channels is 3, and the number of image categories is also 7.

2.The size of the dataset

To facilitate data analysis, balanced control was applied to the seven types of motion data, generating 2,250 GAF images per motion for a total of 15,750 image data points. These data were randomly allocated to the training, validation, and test sets at a 3:1:1 ratio. Specifically, 1,350 images were randomly selected as the training set, 450 as the validation set, and 450 as the test set for each motion. Ultimately, the training set contained 9,450 images, while the validation and test set each contained 3,150 images, with an equal number of images for the seven motion types in each set.

3.Use of dataset

(1) Human activity recognition based on VGG16, EfficientNet-B0, ResNet50, ConvNeXt-Tiny network and ConvNeXt-Tiny transfer learning:

Firstly, the motion signal of one-dimensional time series is converted into two-dimensional image by GAF algorithm, and then five kinds of networks such as ConvNeXt-Tiny are used to recognize and classify the two-dimensional image data of seven kinds of motion.

(2) Human activity recognition based on LSTM and SVM networks:

LSTM and SVM networks were used to identify and classify the one-dimensional time series data of seven kinds of motion.