Research On Real-time Detection System For Particleboard Surface Defects Based On Deep Learning

China is a big particleboard producer.In recent years,with the widespread application of high-efficiency continuous flat press particleboard production line,the production efficiency of particleboard is greatly improved,and the production of particleboard is increasing year by year,which promotes the development of particleboard industry.In the process of high-speed operation of a particleboard production line of unbroken horizontal press,due to many reasons such as raw material composition,raw material ratio,technology and technology,some products will inevitably have defects and defects on the surface.The defects on the surface of particleboard will affect the quality of the board and is therefore difficult to control the veneering.The weaknesses in the surface of particleboard are not difficult to reduce the rate of excellent particleboard products and increase the rate of defective particleboard products.Up to now,Chinese enterprises rely on human visual recognition,detection and grading of drawbacks.There are three disadvantages in this traditional method.On the one hand,under the noisy,dark and complex production environment,workers can observe the particleboard on the high-speed production line with their eyes for a long time,which will soon form visual fatigue,which will lead to missed inspection and false inspection,resulting in the decline of the quality of particleboard.On the other hand,the speed of artificial optical inspection is lower than that of particleboard,and particleboard needs to be slowed down after it is out of the grinder.The low-speed operation mode combined with simulated visual inspection can seriously decrease the delivery efficiency of particleboard.On the other hand,in the industry standard-forestry(CN-LY)standard of"marine veneered particleboard.Definition and classification",the single area of particleboard surface defects shall not exceed 10mm~2,and the area weaknesses of 5-10mm~2 shall not exceed 2,which is hard to catch by mortal eyes.Therefore,the research and development of an accurate and efficient real-time detection system for particleboard surface defects is an effective measure to alleviate the dilemma of China's high-yield particleboard production industry,and also an urgent need of global particleboard enterprises.In this paper,based on the real-time target detection algorithm of deep learning Yolo V3,a compact Yolo net network model is proposed and constructed,and the network model is trained with a lot of data.Then the hardware system of particleboard defect detection is built-in the production line at the factory,and the network model is optimized.The field test shows that the algorithm network model of real-time detection of particleboard surface defects proposed in this paper is accurate for the classification and grading of superior boards,defective boards and protection boards,and it is real-time and efficient for the detection of various common defects such as large particleboards,rubber spots,oil stains,looseness,sand leakage,cracks and edge chipping.The main research contents and conclusions are as follows:1.Through reading a large number of literature,practical research,summed up the status qua of research at home and abroad.Generally,there are two kinds of commonly used target detection algorithms,one is a customary feature target detection algorithm,the other is profound learning model target detection algorithm.2.This paper introduces the basic structure and principle of convolutional neural network,and the elementary principle of target detection algorithm of Yolo V3 model.According to the actual requirements,according to the micro characteristics of particleboard surface weaknesses,production line detection real-time and other requirements.This paper presents a compact Yolo net network model,which uses the pure collected image of particleboard surface in the factory production line to establish training and testing sets for model training and testing.3.Depending on the material of the target particle board,the testing place is the production line of the factory particle board,and the testing index is required.The project team designed a real-time detection system of particleboard surface drawbacks based on meaningful learning.The hardware composition and selection of the system design scheme,the system control scheme and the design of human-computer interface software are introduced.The design process and function of the whole system is laid down in detail.4.The system trial operation was carried out in the particleboard production plant of Fenglin Yachuang wood based panel Co.,Ltd.In Huizhou City,Guangdong Province.The real-time detection system of particleboard surface defects based on arcane learning was installed and debugged in the production line.After one month of real-time detection,the model parameters were online optimized,and the test results of plate classification and classification fully met the needs of the enterprise.The real-time detection system of particleboard surface defects based on subconscious learning developed by our research group can realize the functions of real-time acquisition of particleboard surface image,fast training and detection model,real-time intelligent recognition,board grading and classification,accurate and efficient warehousing.The warehousing time of each particleboard is only 253ms,and the classification accuracy rate is 97.7%.It can improve the effectiveness of board grading,classification and warehousing It can meet or even exceed the national standard of particle board surface detection accuracy,and promote the progress of intelligent automation of particle board production line.