Objective: This study introduces a novel deep-learning-based orientation recognition approach for detecting intraoperative lung orientation during robot-assisted anatomical resections, including lobectomy and segmentectomy. This method can potentially aid in anatomical structure identification, facilitate training and education, improve procedural efficiency, and enhance intraoperative imaging navigation. Methods: We developed a unique dataset encompassing various pulmonary procedures, being the first to report on recognition of intraoperative orientation. The TeCNO model, initially developed for laparoscopic cholecystectomies, was adapted for this study. Model performance was evaluated using accuracy, precision, recall, and F1-score, and we explored the influence of dataset composition, intraoperative factors such as 3D model presence, and visual impairments. Results: The model achieved an overall accuracy of 70%, indicating potential in recognizing lung orientation. High performance was achieved in recognizing non-surgical sequences, ‘Fissure’, and ‘Inferior’ views. ‘Posterior’ and ‘Anterior’ views showed inferior performance. Variability in performance was attributed to the heterogeneity of orientation transitions and increased complexity compared to more standardized procedures. The limited dataset size and imbalances in label distribution potentially impacted model performance. Conclusion: This study demonstrates the feasibility of applying phase recognition to detect orientation of the lung and exploring how the unique characteristics of our dataset affect model performance opposed to surgical phase recognition. The results suggest promising applications for intraoperative imaging guidance and automated adjustment of 3D models, particularly for complex orientations like the interlobar ‘Fissure view’. Future research should focus on enhancing model performance and assessing its clinical implementation in diverse surgical settings.