Deep learning has significantly improved the accuracy and efficiency of pest detection from images. From CNN-based models to transformer-based architectures like ViTs and BEiT, researchers continue to innovate, addressing challenges, and enhancing interpretability. As the field advances, the integration of diverse data sources and real-time monitoring systems holds the potential to revolutionize precision agriculture. Pest detection from images plays a pivotal role in modern agriculture and pest management. With the increasing demand for efficient and sustainable agricultural practices, early identification and mitigation of pest infestations are crucial. A BEiT-based deep model is designed and experimented for agricultural pest detection from images. The proposed BEiT based deep model utilizes an efficient image transformer architecture to achieve optimal performance on agricultural pest image classification. The proposed model has been evaluated against traditional deep learning models – ResNet-50v2, EfficientNetv2L, InceptionResNetv2, DenseNet. Through experimental evaluation, it is proven that the proposed BEiT-based model performed agricultural pests detection effectually by demonstrating better performance of 3–5% than other considered models. BEiT’s global context understanding and learning capabilities make it a promising candidate for agricultural pest detection, enabling accurate recognition of pests with complex visual patterns and the ability to generalize across diverse agricultural environments.