Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder in which individuals often face social difficulties as well as language and communication challenges. Micro-expressions are extremely brief changes in facial expression. Moreover, the micro-expressions exhibited by individuals with ASD frequently represent an accurate reflection of their internal feelings. Therefore, using the Cinemetrics method to extract micro-expressions from ASD patients in movies and targeting them for detection can help doctors make early diagnosis of ASD patients. In this paper, we establish a dataset of micro-expressions of ASD patients in movies, use the improved YOLOv8-SMART algorithm for target detection, and compare it with other target detection algorithms without improvement. The comparison results prove that our algorithm effectively improves the recognition of micro-expressions, which provides reference value for future practical applications in the task of micro-expression recognition in ASD patients.