This research article proposes a physical education structure that takes the cultivation of students’ core competences as the center. This research carries out a comparative teaching experiment by taking two basketball elective classes of the 2023 grade in NJ University as the research objects. For the further bettering of teaching content that physical education teachers use, this article designs a time-space obtaining method for basketball movement postures which is on the basis of FSW, which combines BiLSTM and Seq2Seq to establish the BiLSTM-S2S action feature recognition model. It was found that the average recognition rate of the BiLSTM-S2S model on the eight basketball action gestures was 44.54%, The accumulative score of the experiment class students’ basketball movement abilities has achieved 78.65±2.12. This score is 71.05% bigger than the score which belongs to the students in the control class. Furthermore, when put into comparison with the control-class students, there existed an obvious promotion in the core abilities of the experimental-class students (P<0.01).
