MSYM: A Lightweight Yolo-Mamba Network for Plants Recognition in River and Lake Riparian Zones

Plants are an important ecological component of river and lake riparian zones, and the monitoring of plants growth is necessary for the protection of water environment. The traditional manual inspection mode of riparian zone plants is time-consuming and laborious, and the use of artificial intelligence inspection mode has the advantages of high efficiency and real-time. However, for engineering application scenarios such as unmanned aircraft and unmanned boats, the current methods for plant identification in river and lake riparian zones can hardly meet the edge computing deployment requirements such as high accuracy and low computation capacity at the same time. In this study, we constructed a lightweight multiscale Yolo-Mamba network, called MSYM. MSYM's multiscale channel backbone network realizes selective feature combination and convolution (MCC), and improves the information interactivity between feature channels. In addition, the multi-channel visual state space block (MVSS block) is designed to superimpose and fuse multiple scanning modes to improve the efficiency by adopting the multi-channel scanning mode for the computational complexity requirement. At the same time, we constructed a sample bank of plants in the riparian zones of rivers and lakes, which is based on the spatial and temporal patterns of plants in the riparian zones of rivers and lakes, and a sample bank of plants in the riparian zones of rivers and lakes, which is based on the coupling of watersheds with climatic patterns. The experimental results demonstrate that the proposed model achieves excellent recognition accuracy under the application environments of unmanned aerials photography and unmanned boat cruising, with Precision\Recall\F1\IoU reaching 0.947\0.872\0.908\0.831, and FLOPs reduced from 12.1 by yolov8n to 4.1.