Plant Identification in Sanjiangyuan Region Based on Improved Yolov8n

As a crucial ecological security barrier in China, the protection and management of plant resources are of paramount importance. However, there are limitations in the application of existing plant identification techniques in the Sanjiangyuan region. We have constructed a dataset containing 26,340 plant images, which enhances the richness of the data and effectively reduces the risk of model overfitting. Based on this, this paper proposes a plant recognition algorithm based on an improved YOLOv8n architecture: GAM_Attention is introduced to fuse channel and spatial information; C2f_DCN is used to flexibly adjust convolutional sampling positions; and SPPF is replaced by SPPFCSPC to extract multi-scale features. The experimental results indicate that the model demonstrates outstanding performance on the self-constructed Sanjiangyuan plant dataset. Specifically, the precision, recall, mAP@50, and mAP@50-95 metrics on the test set have improved by 8.97%, 6.31%, 7.81%, and 10.53 %, respectively.