Abstract: Through the combination of artificial intelligence (AI) and the Internet of Things (IoT), smart agriculture has emerged as a revolutionary step to increase crop yield and sustainability in recent years. This integration has made it possible to continuously monitor farms and automatically assess the health of crops. However, a number of issues plagued the current Convolutional Neural Network (CNN)- based smart agriculture system, including limited applicability in remote agricultural regions, inconsistent data collecting in a variety of field conditions, and poor generalization on single crop diseases. A smart agriculture system based on Vision Transformers (ViT) is suggested as a solution to these problems. The first of four layers in this system architecture is the data acquisition layer, which is equipped with sensor nodes and cameras to gather environmental data and photos of plant leaves. The communication layer follows, which is in charge of carrying out data transfer to the following layer, The processing layer comes next, when threshold evaluation and Simple Moving Average (SMA) filtering are used to preprocess environmental data. Additionally, bilinear interpolation is used to scale and normalize the image data. The pretrained ViT model is then fed this pre processed data in order to classify plants with multiple crops and diseases. Finally, the user receives an interactive web-based dashboard with the output, which comprises the illness kinds that were identified and their confidence levels. The suggested ViT-based system beat the current method and achieved greater accuracy, according to experimental results.

Keywords: Vision Transformer, CNN, IoT, KNN, Simple Moving Average

Downloads: | DOI: 10.17148/IARJSET.2025.12832