Dataset for Human Activity Recognition with FMCW Radar Using Few-Shot learning

 Our dataset has a total of 8 actions, 7 people(P1-P7), and 3 experimental environments(Room-A,Room-B,Room-C). There are a total of 3 directions in each environment, with 5 samples of each action taken for each person in each direction, so the number of samples is 360(samples/person)*7 = 2520.    We use two environments(Room-A and Room-B) and 5 people(p1-p5) as training and validation, 150(samples/action)*8(actions) = 1200,the Classes in front of  the symbol (_), followed by the serial number, The 1-75 samples are taken from the training set and the 76-150 samples are taken from the val set. i. e,the images of action Bow are named 0_1~0_75(train_bow),0_76~0_150(val_bow).    We take the last environment(Room-C) and 2 people(P6 and P7) as a test, 30(samples/action)*8(actions) = 240.Image name naming rules: the symbol (_) in front of the category, followed by the serial number, where the test set of 30 samples per action, each category is sorted according to 1-30, i. e,the images of action Bow are named 0_1~0_30(test_bow).Each action is classified as follows:0 is bow1 is boxing2 is falldown3 is handup4 is stand5 is squat6 is sit7 is walk