Research On Key Technology Of Automatic Measurement Of Wood Volume Based On Deep Learning

With the proposed goals of carbon neutrality and carbon peaking,improving the level of forestry intelligence has become a major support for the dual carbon goals.The diameter class and volume measurement of wood is an important direction of forestry intelligence.The manual measuring method that has been widely used by enterprises for a long time has low efficiency,strong subjectivity and high labor intensity.With the development of image processing technology,the measurement of wood diameter in wood images has become possible.However,in the actual timber yard,for the dense vehicle-mounted timber diameter class measurement,affected by factors such as the blocking of the timber end faces,the mutual adhesion of outline boundaries,and the color deviation of the tree core,the detection rate of timber,the accuracy of outline segmentation,and the efficiency of inspection rulers have fallen into bottlenecks.In addition,at this stage,the State Forestry Administration is actively promoting the development of artificial intelligence in forestry.Therefore,this paper conducts in-depth research on difficult problems such as artificial intelligence image segmentation,wood contour fitting,diameter class and volume measurement of dense vehicle-mounted wood.The main work is as follows:First,make a wood image dataset.The gantry image acquisition system and mobile phone camera are used to collect a large number of timber images at the timber production site of a forestry listed company;Labelme tool is used to complete the timber contour labeling;Based on the CLo DSA library,the synchronous data enhancement of wood images and corresponding annotation information is realized,providing rich image data for model training.Second,the WM R-CNN wood end face instance segmentation model is designed.WM R-CNN is based on the Mask R-CNN algorithm as the basic network framework.In order to improve the basic comprehensive performance of the original algorithm for wood detection,the Res2 Net module and the PAFPN structure were used in the Backbone network to improve the Res Net structure and the FPN structure of the original algorithm.At the same time,in view of the serious problem of missing detection of dense small wood in the original algorithm,a model optimization training method named SWDE for enhanced small wood detection is proposed.Combining the two model improvement and optimization methods,several sets of control experiments were designed for WM R-CNN,which verified the great improvement of WM R-CNN in segmentation accuracy,mask segmentation quality and wood detection rate.On the test set,the true detection rate of wood reaches 98.58%.Finally,a wood volume measurement algorithm based on Aruco QR code is designed.Aiming at the problem that the least squares ellipse fitting algorithm has poor restoration of the incomplete wood contour,a wood contour ellipse fitting algorithm combining invariant moments and the least squares method is designed.Aruco code is used as an auxiliary marker for 2D image wood end face diameter measurement and volume measurement,and the original Aruco code corner detection algorithm is improved based on the grayscale enhancement principle to ensure high robustness to Aruco code corner detection;According to the corner coordinates of the Aruco code,the image measurement parameters are calculated and the perspective transformation of the wood image is realized.Then,the ellipse information of the wood profile is fitted to realize the measurement of the diameter and volume of the wood end face.In this paper,multiple sets of control experiments were designed,and the average volume comprehensive error rate of the double Aruco codes timber volume measurement method was1.14%,which verified the effectiveness and applicability of the method.Compared with the artificial timber diameter measurement data,the experimental results of the designed automatic timber volume measurement system in the timber yard show that the average error of diameter measurement is less than 5 mm,and the error rate of volume measurement is less than 2%.The results of wood diameter class detection and volume measurement meet national standards.and requirements,and the detection efficiency is greatly improved,which proves that the method in this paper has high application value.