DRR-YOLO: A Study of Small Target Multi-Modal Defect Detection for Multiple Types of Insulators Based on Large Convolution Kernel

The existing insulator defect detection algorithms are mainly characterized by their ability to identify only a single type of defect, accompanied by relatively low accuracy, a Dilated Re-parameterized Residual-YOLO (DRR-YOLO) algorithm is proposed, which is capable of identifying four defects of each of the four insulator types. Firstly, a new module DRR is proposed based on Dilated Re-param Block (DRB) and Dilation-wise Residual (DWR), which is combined with the C2f module of the YOLOv8 model to enhance the model’s detection effect for targets at different scales; Secondly, the Large Separable Kernel Attention SPPF (LSPPF) module is proposed to replace the SPPF in the original model, which enables the model to better preserve the feature information of the target while streamlining the structure; In addition, the advantages of MPDIoU and Inner-IoU are combined, and the loss function of the original model is replaced with Inner-MPDIoU. Finally, the DRR-YOLO model was compared with other existing models. Experiments demonstrate that the accuracy of the proposed model is superior to that of other models. Compared with the best - performing YOLOv5s and YOLOv8s, the mAP value on the validation set has increased by 1.9 percentage points and 2.2 percentage points respectively, reaching 94.7%. At the same time, it can be seen from the detection results that the proposed model has the highest confidence in detecting small targets.