Research On Defect Detection Algorithm Of Solid Wood Plate Based On Deep Learning

Chinese wood resources are limited.In order to improve the utilization rate,efficient and reliable detection of wood defects by machine vision can not only eliminate the shortcomings of manual detection,such as high pressure,low efficiency and poor accuracy,but also help to strengthen the intelligent level of wood processing manufacturers.This paper introduces the development status of machine vision detection technology at home and abroad,and discusses the theory and algorithm of wood defect detection based on machine vision.As a human visual simulation,machine vision combined with AI can significantly improve the detection efficiency,reliability and accuracy.It is not only conducive to promoting the transformation and upgrading of wood processing industry,but also conducive to strengthening the automation and intelligent development of related companies.It is of great value not only for the research of management theory but also for practical application.Therefore,this topic based on deep learning of solid wood defect detection algorithm research,including the following aspects:Firstly,the image acquisition device is constructed to collect the surface defect images of solid wood mainly made of Chinese fir,and store them in the sample database.The defect types focus on live / dead joints,cracks,etc.in order to ensure the accuracy and rationality,the deep learning algorithm is selected to complete the defect detection.Some classical target detection algorithms such as R-CNN,Fast R-CNN,Faster R-CNN,Mask R-CNN are studied.Through the comparative analysis of advantages and disadvantages,the paper considers the actual situation,Mask R-CNN algorithm is finally selected as the research model of this topic.The neural network architecture search algorithm is introduced.Glance Network is proposed as the front-end network of the model,and the NAS-Mask R-CNN model is built at the same time.Starting from the image of wood surface defects,using the NAS-Mask R-CNN algorithm for defect detection and result analysis,it is found that the NAS-Mask R-CNN algorithm has a good performance in detection speed and detection accuracy.In order to further optimize the model,a confidence optimization strategy is introduced to improve the stability of the algorithm in engineering applications.At the same time,a multichannel selection optimization strategy is proposed to optimize the network structure.After comparing the experimental results,it is found that the improved NAS-Mask R-CNN algorithm has greatly improved the detection speed.Through the completion of a series of work such as the construction of the experimental platform and the development and optimization of the experimental algorithm,the finally proposed algorithm has reached the project requirements in terms of detection accuracy and detection time.