Acta Scientiarum Polonorum Technologia Alimentaria

Computer vision has become a cornerstone technology in the food and agriculture industries, driving innovation and enabling automation across a wide range of processes. Within this field, object detection plays a critical role, supporting efficiency, accuracy, and scalability in real-world applications. The Transformer, first introduced in natural language processing, demonstrated outstanding performance thanks to its powerful self-attention mechanism and parallel processing capabilities. More recently, it has been rapidly adopted in object detection and is emerging as a strong alternative to traditional convolutional neural networks. However, much of the related research remains scattered and interdisciplinary. This paper systematically reviews the development of transformer-based models for computer vision, analysing research trends, key topics, and distinctions from other algorithms. It introduces the basic architecture of the Vision Transformer (ViT) and other transformer-based vision models, explains core principles such as self-attention and multi-stage processing, and examines applications in food and agriculture, including food quality analysis, crop monitoring, pest and disease detection, and weed identification. Challenges and future directions of transformer-based models are also discussed, alongside a review of the latest research for reference. By consolidating a large body of literature, this study provides a comprehensive overview of the structure, development,  dvantages, and limitations of transformer-based vision models, while highlighting their potential to deliver more intelligent, sustainable, and efficient decision-support systems for precision food and farming practices.