In cybersecurity education, the assessment of penetration testing experiments mainly relies on the instructor's on-site observation and result verification, which make it difficult to track students' operational paths. In recent years, AI-based video behavior recognition technology has been used to analyze experimental operation paths, but it has high demands for model training, which limits its use on large-scale application scenarios. This paper proposes a lightweight screen recording analysis method in the cyber range, combined with the use of video frame extraction, OCR command recognition, and Levenshtein distance algorithm, to analyze and evaluate the experimental process of penetration testing. The results demonstrate that the method can quantify student performance. For instance, a similarity ratio of 0.40 yields a score of 4/10, indicating missing or incorrect steps in the penetration process. These findings suggest that this method improves learning feedback and provides a scalable approach to process-oriented evaluation in cybersecurity education.