Research On Classification Algorithm Of Finger-Plate Based On Deep Learning

With the improvement of people ’ s quality of life,people ’ s demand for solid wood furniture is also increasing year by year.However,the per capita share of forest resources in China is low,and the utilization rate of wood is low.Therefore,the state advocates the use of intelligent and automatic wood processing equipment to improve the utilization rate and quality of wood.At present,due to technical and economic factors,most of China ’s finger-plate processing enterprises still use manual detection method to complete the classification of fingerplates.This method leads to low degree of automation of finger-plate processing machinery,inaccurate detection results,sawing and classification affected by subjective factors of workers,and lack of scientific and objective.In view of the above problems,this paper proposes a classification system of the fingerboard based on deep learning,and completes the following main research work :Firstly,the surface characteristics of 14 types of finger plates are analyzed by referring to relevant references and combining with the actual classification standards of finger plate processing enterprises.Design and build a fingerboard classification system platform.Secondly,the basic composition of image classification algorithm based on deep learning is analyzed.The data enhancement and label smoothing technology are used to improve the generalization ability of the model,and the model lightweight criterion and efficiency improvement method are used to optimize the reasoning speed of the model.By using data enhancement method to enhance the training set,it is verified that improving the diversity of training set samples can effectively improve the accuracy and generalization ability of the network model.The effectiveness of the lightweight network model design criteria proposed by Shuffle Net V2 is studied and verified,which provides a theoretical basis for the lightweight of subsequent models.The label smoothing technology is proposed to solve the over-fitting problem of the model,and the experiment verifies that label smoothing is conducive to improving the generalization ability of the model.The experimental results show that the use of GPU acceleration will greatly improve the training and reasoning efficiency of the model.In terms of the classification of the fingerboard,considering the classification accuracy and classification efficiency of the model,we extracted the feature extraction skeleton CSPDark Net of YOLOV5 s with small network depth and width,and added the global average pooling,full connection layer and Softmax layer to study on this basis.Aiming at the complex surface texture features of fingerboard samples and the confusion between categories,ECA＿CSPDark Net combined with channel attention mechanism is proposed to ensure the information interaction between channels and improve the classification accuracy of the model.According to the lightweight design criteria of the model,the channel separation technology is used to reduce the memory access cost of the model,improve the reasoning speed of the model,and scale the depth of the improved model.Experiments show that the classification accuracy of the improved network model on the 14 types of fingerboard sample data set reaches 92.24 %,and the speed of image classification is also improved.Finally,in order to better apply the fingerboard classification algorithm to the actual production,the software design and function realization of the classification system are carried out through the Py Charm software development platform.The effectiveness of the classification system designed in this paper in the classification task of the fingerboard is verified in practice.