Sugarcane yield prediction using NOA-based swin transformer model in IoT smart agriculture

Abstract

The Internet of things (IoT) empowers precise organization and intelligent coordination for industrial facilities and smart farming, enhancing agricultural efficiency. Sugar production relies on various auxiliary elements, but in laborintensive smart agriculture, creating accurate forecasts is a formidable challenge. Machine learning emerges as a potential solution, as current convolutional neural network-based phase recognition techniques struggle with longrange dependencies. To address this, a temporal-based swin transformer network (TSTN) is introduced, comprising a swin transformer and long short-term memory (LSTM). The swin transformer employs attention mechanisms for expressive representations, while LSTM excels at extracting temporal data with long-range dependencies. The nutcracker optimizer algorithm (NOA) fine-tunes LSTM weights. TSTN effectively blends these components, providing spatiotemporal data with enhanced context. This model outperforms competitors in accuracy, as demonstrated through testing with data from Uttar Pradesh. The integration of IoT and TSTN marks a significant advancement in optimizing agricultural operations for increased productivity and efficiency. In the comparative analysis, the proposed TSTN-NOA model achieves better performance and results than other existing models. ARTICLE HIGHLIGHTSSugarcane Yield prediction is carried out by Temporal-based Swin Transformer Network (TSTN), because existing CNN models are not very adequate for the extraction of long-range dependencies. The Swin Transformer incorporates the attention mechanism to encode remote dependencies and learn highly expressive representations. The proposed model is focused on Uttar Pradesh mainly because sugarcane was cultivated on 28.53 lakh hectares in Uttar Pradesh, which is more than any state in the country. To improve the classification accuracy, the Nutcracker optimizer algorithm is used for selecting the weight of the Long Short-Term Memory that is one of the components in the TSTN model.