Research on wood defect detection model Wood-Net based on YOLOv7
In order to improve the low efficiency and high labor cost of manual wood defect identification,and to achieve rapid and accurate detection of different defects during wood processing for enhancing wood utilization,this study addresses the limitations of existing object detection networks in wood defect detection,such as low detection accuracy,high false positive rate,and limited recognition categories.A deep learning network for wood defect detec-tion named Wood-Net was designed.Wood-Net incorporated the efficient channel attention(ECA)mechanism into the backbone network of YOLOv7 to better differentiate subtle differences among wood defects.The combination of ECA and Res2Net formed the ECA-Res2Net module,which overcame the insufficient inter-channel communication capability of Res2Net and enhanced the network's ability to extract more fine-grained features.The ECA-Res2Net module was then combined with the spatial pyramid pooling and channel spatial pyramid convolution(SPPCSPC)in the ResSPPCSPC module,which increased the descriptive capacity of the image's own features,resulting in the novel Wood-Net method.Precision,recall,mAP@0.5 and mAP@0.5:mAP@0.95 were used as evaluation metrics for the system performance.Wood-Net was trained using a self-built dataset,and experimental data was obtained.The results showed that the Wood-Net model achieved a 4.52％improvement in precision and a 6.62％improvement in mAP@0.5:mAP@0.95,compared to the baseline model YOLOv7 in wood selection.It also outperformed the baseline model YOLOv5s with a 6.79％improvement in precision and a 5.67％improvement in mAP@0.5:mAP@0.95.The ECA attention mechanism effectively enhanced the inter-channel information interaction of E-ELAN.The Res2Net module had strong capability in extracting fine-grained features,and its introduction into the network im-proved the convergence speed of various performance indicators.Addition of ECA to Res2Net enabled the considera-tion of inter-channel relationships and accomplishes information fusion,leading to improved detection performance.
国家科技期刊平台
NSTL
万方数据
维普
