Attention Based GC-GhostNet for Forest Pests Detection

Forest pests are very harmful to forestry protection. In order to facilitate the monitoring and research of forest pests, this paper proposes the GC-GhostNet model. This model is based on GhostNet and adds a global attention mechanism Global Context block module as the backbone of the network. After extracting the feature maps of different dimensions, the model uses the BiFPN structure to fuse the features of 3 different dimensions, and finally predicts the category and the bounding box respectively. Experimental results show that our model reaches 84.3 mAP@0.5 on Baidu’s AI insect recognition data set, which is 5.2% higher than GhostNet. Compared with other models, it also performs well in terms of accuracy and detection speed. Applying this model to the forest management process can better detect pests and managers can take protective measures in time.