Object Detection Model of Cucumber Leaf Disease Based on Improved FPN

Accurate identification of crop diseases was particularly important for disease prevention and control. This paper improved the traditional FPN and proposed a new feature fusion structure SE-FPN. The multi-receptive field feature enhancement module was used to expand the receptive field and global understanding ability of high-level features, and the semantic activation module was used to activate low-level features through high-level features. To enhance its semantic information, this paper proposed a cucumber leaf disease object detection model based on SEFPN module and transfer learning. On the self-collected dataset, the mAP of the model in this paper reached 86.1% Under the same experimental conditions, compared with the traditional FPN and PAFPN structures, the experimental results showed that the improved feature fusion structure was better than the structure before improvement, the mAP increased by 3.5% and 2.9%, respectively. The model in this paper had high detection accuracy, and good detection results for complex background environment, leaf occlusion and too small leaves, which provided a reference for accurate detection of crop diseases.