A Novel DME-YOLO Structure in a High-Frequency Transformer Improves the Accuracy and Speed of Detection

Abstract

Traditional YOLO models face a dilemma when it comes to dim detection targets: the detection accuracy increases while the speed inevitably reduces, or vice versa. To resolve this issue, we propose a novel DME-YOLO model, which is characterized by the establishment of a backbone based on the YOLOv7 and Dense blocks. Moreover, through the application of feature multiplexing, both the parameters and floating-point computation were decreased; therefore, the defect detection process was accelerated. We also designed a multi-source attention mechanism module called MSAM, which is capable of integrating spatial information from multiple sources. Due to its outstanding quality, the addition of MSAM as the neck of the original YOLOv7 model compensated for the loss of spatial information in the process of forward propagation, thereby improving the detection accuracy of small target defects and simultaneously ensuring real-time detection. Finally, EIOU was adopted as a loss function to bolster the target frame regression process. The results of the experiment indicated detection accuracy and speed values of up to 97.6 mAP and 51.2 FPS, respectively, suggesting the superiority of the model. Compared with the YOLOv7 model, the experimental parameters for the novel DME-YOLO increased by 2.8% for mAP and 15.7 for FPS, respectively. In conclusion, the novel DME-YOLO model had excellent overall performance regarding detection speed and accuracy.

MCAW-YOLO: An Efficient Detection Model for Ceramic Tile Surface Defects

“…This model utilizes the K-means clustering algorithm to optimize the size of anchor points and incorporates the efficient channel mechanism in the backbone network to enhance feature representation. Kang et al [15] introduced the DME-YOLO model for defect detection in the appearance of high-frequency transformers. By reusing features, they accelerated the detection speed.…”