Recognition of Plantation Farming Behavior Based on Deep Learning

Rice and other crops' growth relies on proper farming, such as plowing and pesticide use, which are critical for yield and food safety. Plantations often manage farm activities through manual recording of farming behaviors. To address the inefficiency of manual recording, this study introduces the ED-FBNet model, based on the X3D method. It streamlines the model with pruning for edge device compatibility, enhancing performance with the ECA attention mechanism. Results show the ED-FBNet's mAP score at 93.19%, outperforming models like FWNet, Slowfast, and X3D by 3.25%, 6.37% and 8.92% respectively. Successfully deployed on-edge devices, this marks a significant step towards smart, automated agricultural production and agricultural product traceability.