Research Of Improved Mask R-CNN-Based Detection Segmentation Algorithm For Wood Defects

Wood is an important natural resource,but defects in wood can seriously affect its commercial value.To improve the utilization of wood,defects need to be detected and split during the processing of wood.Manual visual inspection and traditional sensor inspection can no longer meet the needs of rapid industrial defect detection.Deep learning algorithms and machine vision technologies have made great progress in the field of defect detection in recent years.Mask R-CNN(Mask Region-Based Convolutional Neural Networks),as the best achievement in the field of deep learning,can achieve both defect detection and localization and contour segmentation.Therefore,this paper proposes a new detection segmentation method for wood defect s based on the Mask R-CNN algorithm with adaptive improvement.The main research contents are as follows:Firstly.The collected wood defect samples are statistically analyzed,and the Cycle GAN(Cycle Generative Adversarial Network)is adopted to exchange the texture color information of the three types of tree species and generate new defect images to solve the problem of uneven distri bution of wood defect categories.Various online data augmentation schemes are adopted to solve the problem of insufficient wood defect samples and insufficient network robustness.The experimental analysis of comparison with the model before data processing shows that the model detection precision is improved by 6.5% and segmentation precision is improved by4.1%.Secondly.To address the problem that the residual network has insufficient ability to extract contextual information,the residual network is improved,and the layered residual module is established to extract feature information of multiple sensory fields at each layer of the network.To address the problem that the conventional convolution is inadequate for fitting irregular defects,the deformable convolution is adopted to replace the conventional convolution structure to improve the model's ability to adapt to irregular geometric deformation defects.The experimental analysis was conducted in comparison with the original model,and the detecti on precision was further improved by 3.4% and segmentation precision was further improved by 1.0% using the data processed dataset.Lastly.Comparing this research algorithm with classical object detection algorithm,classical instance segmentation algorit hm,and other wood defect detection algorithms in terms of defect detection precision,segmentation precision,and testing effect,our algorithm performs best in all dimensions.