This paper discusses the development of school soccer with the help of artificial intelligence. Propose a machine learning-based action feature extraction method for students in school soccer. Obtain action images of students playing soccer and identify the actions of students in school soccer based on the threshold recognition algorithm. The Harris 3D operator is used to establish the potential function of the action sequence, and based on the potential function of the action sequence, the AdaBoost algorithm is used to filter the action feature data of the students in soccer, which is used as the training sample to realize the action feature extraction of the students in soccer. To extract the effective feature values and improve the recognition accuracy of the algorithm, a soccer action recognition model based on SVM was constructed. The feasibility of the DTW scoring method in the field of soccer action recognition has been verified. The SVM algorithm model has the strongest denoising ability, and its feature action recognition rate is maintained between 80% and 90% and the recognition rate of features with large action amplitude is higher, which is suitable for the recognition of soccer actions in this study.