IAE-SDNet: An End-to-End Image Adaptive Enhancement and Wheat Scab Detection Network Using UAV

Wheat scab is a severe fungal disease caused by Fusarium. The precise detection of scab is crucial for improving the wheat yield. The utilization of RGB images captured by unmanned aerial vehicle (UAV) can effectively enhance the wheat scab detection efficiency. However, UAV images are influenced by environmental conditions. In addition, the scab in the image is extremely small and the background is complex. Therefore, to boost the wheat scab detection accuracy, this study proposes an end-to-end image adaptive enhancement and wheat scab detection network (IAE-SDNet). In the head of the network, an image adaptive enhancement (IAE) module aims to enhance the quality of UAV images and performs end-to-end learning with the transformer-based object detection module SDNet. A cascading inverse residual module (CIRMB) is designed to boost the feature extraction capacity for the disease area. A deformable multihead attention encoder (DMHA-Encoder) is deployed to augment the semantic information of the advanced features of wheat while maintaining the representation ability of global attention. A multiscale feature fusion block (MFFB) is designed to effectively fuse wheat disease features of different scales, reducing the impact caused by the inconsistency in size between the complex field background and the disease area. Finally, to optimize the detection of small lesion areas, an MNIoU bounding-box regression loss function is proposed. The experiment indicates that this method has improved average precision (AP) by 6% and the Recall by 5% compared with the baseline network. This study provides a feasible approach for the precise detection of wheat scab, with high application prospects.