An Improved YOLOv11 Method With Mamba Module for Multiscale Tea Disease Detection

Deep network–based models have been widely applied in smart agriculture, including tea disease detection tasks. However, the inference of existing deep models requires a large amount of computing resources and storage space, making it difficult to effectively deploy on mobile devices. In addition, existing deep models have low performance for multiscale object detection. To this end, this paper proposes a variant YOLOv11 model based on the Mamba module for tea disease detection. Specifically, in the existing YOLOv11 model, we replace the self‐attention mechanism with the Mamba module to achieve long‐range dependency modeling with linear complexity. To further improve the performance of our proposed YOLOv11, we introduce the normalized Gaussian Wasserstein distance (NWD) loss for alleviating changes in the shape and scale of tea disease targets. To verify the effectiveness of our proposed model, we conducted a large number of comparative experiments on a public dataset and our self‐built tea disease dataset, and all the results confirm that our proposed model achieved more competitive performance in terms of accuracy and speed.