Research On Log-End Area Recognition Methods

Wood counting and volume calculation are important data of log production and sales.At present,the measurement methods of wood volume include manual measurement and laser measurement.These methods consume manpower and material resources,cost is high,accuracy is low,and can not meet the requirements.In order to solve the problem of log number statistics and volume calculation in the natural environment,the paper collected the log pile images under different storage environment and different shooting conditions,and used computer vision technology to effectively locate and identify the logs end.1.Image segmentation based on color difference clusteringAiming at the problem that the existing segmentation methods are not accurate in the complex natural environment,this paper proposes a log image segmentation algorithm based on color difference clustering,which uses R-G,G-B and B-R as clustering features to cluster the background,logs end and pores unsupervised,extract the threshold value,and realize the image segmentation.Compared with Otsu's method and color difference method,it has achieved better results.2.Log section detection based on step by step operationCombined with the open operation and watershed algorithm,a log section detection algorithm based on the step-by-step open operation is proposed.Firstly,the adhesion is removed by the change of the radius of structural elements,and then the target area with similar size of connected area is segmented by watershed algorithm.Finally,the logs end recognition and counting statistics are realized.In the natural environment,the positive rate was 91.88%,the false rate was 5.08%,and the missed rate was 8.12%.3.Logs end recognition based on deep learningAiming at the problem of low detection rate of traditional methods,this paper makes an in-depth study on logs end recognition based on deep learning.Firstly,the target of the sample image is labeled,and the rectangular annotation frame is established to divide the background and logs end into two types.Using the single shot multi box detector(SSD),a regression based deep learning framework,a basic network is established to integrate the feature images,and then six different convolution kernels are used to extract the features to realize the target recognition of different sizes.The accuracy of the experimental results is 94.87%,the recall rate is 91.34%.The experimental results are verified by taking pictures in Chengde Weichang forest farm.The computer vision technology can better solve the identification of logs end in the natural environment,which provides a theoretical basis for the subsequent calculation of log volume.