Research on Stored-Grain Pest Detection Algorithm Based on Improved RT-DETR

It is of great significance to conduct efficient and accurate pest detection during grain storage. To address the problems of low detection accuracy and frequent false or missed detections encountered by traditional algorithms under complex backgrounds, an improved RT-DETR detection model is proposed in this paper. Based on the existing multi-scale enhancement module and gated partial attention mechanism, a Multi-scale Enhanced Gated Partial Attention Network (MEGPANet) is constructed to replace the original backbone network, thereby enhancing the model’s ability to extract pest edge contours and local features. A Diverse Branch Block (DBB) is introduced to replace the original RepBlock module, strengthening the model’s cross-scale feature fusion capability. A Conditional Position Encoding Generator (PEG) is incorporated into the Attention-based Intra-scale Feature Interaction (AIFI) module to dynamically inject positional information, further improving the localization accuracy and detection precision of tiny pests. Experimental results reveal that the precision, mAP@0.5, and mAP@0.5:0.95 of the improved model are increased by 0.9%, 5.5%, and 3.4%, respectively. While improving detection accuracy, the computational complexity of the model is also significantly reduced. Compared with other mainstream object detection models, the improved RT-DETR model proposed in this paper shows certain advantages in detection accuracy and overall performance, indicating that the model can better adapt to pest detection tasks in complex stored-grain scenarios.