Objective To realize precise, swift, and non-invasive detection of pork freshness in large-scale cold meat industry chains based on computer vision technology.Methods An image recognition algorithm for pork freshness is proposed based on YOLOv8n. Various data augmentation methods are employed to enhance the pork feature extraction from images. The transfer learning experiment method is utilized, an appropriate optimizer is selected, and the training weights of the model are improved for higher accuracy in the final identification. Based on the YOLOv8n image recognition algorithm, the improved YOLOv8n-cls model is developed by data augmentation and optimizer improvement for the algorithm.Results After transfer learning and improving the optimizer, the average recognition accuracy, recall rate, and mean average precision (mAP) of pork freshness image recognition achieve 99.4%, 83.8%, and 91.4%, respectively, at an image recognition frame rate of 149 Hz, demonstrating promising experimental outcomes. Following normalization training and ablation testing, the accuracy of pork freshness image recognition increases by 0.5% to reach 99%.Conclusion The improved YOLOv8n-cls model improves image recognition accuracy while maintaining requisite speed, meeting demands for pork freshness real-time detection, and recognizing in practical production settings.