Anomaly detection in cropland monitoring using multiple view vision transformer

In recent times, the importance of low-altitude security, especially in agricultural surveillance, has seen a remarkable upswing. This paper puts forward a novel Internet of Drones framework tailored for low-altitude operations. Anomaly detection, which is pivotal for ensuring the integrity of the entire system, poses a substantial challenge. Such anomalies can range from unpredictable weather patterns in farmlands to unauthorized intrusions. To surmount this, a comprehensive deep learning pipeline is proposed in this study. It deploys a vision transformer model featuring a unique attention mechanism. The pipeline includes the meticulous collection of a vast array of normal and abnormal farmland images, followed by preprocessing to standardize data. Anomaly detection is then carried out, and the model's performance is evaluated using metrics like sensitivity (92.8%), specificity (93.1%), accuracy (93.5%), and F1 score (94.1%). Comparative analysis with state-of-the-art algorithms reveals the superiority of the proposed model. In the future, this study plans to explore integrating data from thermal, infrared, or LIDAR sensors, enhance the interpretability of the vision transformer model, and optimize the deep learning pipeline to reduce computational complexity.