Recent advances in Transformer technology for agriculture: A comprehensive survey

Intelligent agriculture is critical for guiding agricultural production and enhancing efficiency through early disease diagnosis, yield estimation, automatic harvest, and postharvest efficient treatment. The conventional methods, including manual, image processing, and CNN (convolutional neural network), have some shortcomings of high labor consumption, subjectivity, poor robustness, and low efficiency. Transformer, one of the latest technological advances in deep learning, has gained widespread adoption in agriculture since its universal modeling capabilities. This paper is the first comprehensive survey of the recent advancements in Transformer- based models within the agricultural domain. Six research questions are proposed and addressed by reviewing relevant literature from different aspects. Two types of Transformer-based models (pure and hybrid Transformers) are reviewed to outline the architecture of Transformer-based models adopted in agriculture. And different applications of Transformer-based models in agriculture are summarized to display the current development of Transformer in agriculture. It also highlights the main challenges faced by Transformer technology in agriculture and discusses the future directions for its application in agricultural sector. This survey is expected to leave readers with deeper thoughts about Transformer-based models in agriculture and help them perform in-deep explorations on Transformer-based models for agricultural applications.