Research On Typical Defects Detection Technology Of Sawn Timber For Automatic Sawing

Sawn wood chip processing is an important part of the wood processing industry,and its production efficiency and quality directly affect the cost-effectiveness of wooden furniture.At present,our country is facing the dilemma of poor forest resource endowment and low comprehensive utilization rate of sawn timber.In order to adapt to the new supply-demand relationship in the domestic sawn timber market,the priority of the domestic sawn timber processing industry must be changed from resourceconsuming to efficient production capacity.Specifically,it is mainly necessary to rely on technological innovation in the sawn timber processing industry to achieve automated production.However,the efficient and high-quality processing of sawn timber is limited by the automatic detection of wood defects.Scholars have carried out relevant research on wood defect detection technology and have achieved good progress,but there are still challenges in practical application:(1)The defect feature extraction method based on traditional digital image processing can achieve no-omission segmentation of defects,but it cannot take the task of defect classification into account;(2)The target detection algorithm based on deep learning can realize the integration of defect detection,classification and positioning while maintaining a high detection speed,but there are problems of inaccurate defect positioning and missing detection at the same time.Aiming at the problem of automatic wood defects detection that restricts highspeed automatic sawing of wood,this paper studies a high-precision real-time defect detection algorithm model,and realizes sawing wood defect detection for automatic sawing.The main contents include the following aspects:1.Establish a dataset of typical defects of sawn timber for automatic sawing.Considering the current situation that there is no high-quality benchmark in the field of sawn timber defect detection and the demand orientation of current sawn timber processing industry,the sawn timber defect dataset in this paper takes rubber wood sawn timber as the research object and a total of 5216 high-quality defect sample images are collected.The images contain 4 classes of typical defects,including dead knot,intergrown knot,shake and inbark.Meanwhile,according to the difficulty of defect detection,the images are divided into three categories,namely simple samples,general samples and complex samples.So far,a quantitative balance is achieved in each class or category of the sawn timber defect images.Above,the dataset in this paper is practical,typical,diverse and balanced.2.Propose a sawn timber defect detection algorithm based on Position SSD(PSSD).Comparing with the benchmark SSD model,the proposed PSSD introduces Io U confidence,EIo U-NMS strategy and Focal-EIo U loss function to solve the problem of the sawn timber defect detection,including repeated-detection and inaccuratepositioning-prediction boxes.As a result,the m AP is increased to 89.24%,and the detection speed reaches 42 FPS.3.Propose an enhancement algorithm of sawn timber defect based on edge extraction.This paper takes advantage of the target location of digital image processing and combines the universal characteristics of sawn timber defects to design an efficient and concise algorithm,including denoising filtering,defect segmentation,outlier removal,edge extraction,which effectively avoids the problem of missed detection of defects in the benchmark SSD algorithm,and extracts the precise edge information of sawn timber defects as positioning priors at the same time.4.Design a combined edge-enhanced deep sawing timber defect fusion detection method.In the task of sawn timber defect detection of this paper,the target detection method of deep learning and digital image processing is combined to achieve complementary advantages.Based on the PSSD algorithm model,the detection accuracy m AP is further improved to 91.28% by using the positioning prior.Meanwhile,the detection speed remains at 40 FPS.To sum up,this thesis has successfully completed the task of sawing timber defect detection for automatic sawing.By applying the combined edge-enhanced deep sawing timber defect fusion detection method,the detection accuracy of dead knot,intergrown knot,shake and inbark has reached 94.15%,90.33%,90.10% and 90.54%,respectively.As for detection speed,the proposed fusion algorithm reaches 40 FPS,which satisfies the real-time detection speed requirements of more than 35 FPS in the actual production of sawmills.At the same time,it is 31 FPS faster than the Faster-RCNN model,which is basically the same as the benchmark SSD model and PSSD model.Even though it is not as good as the 58 FPS of the Yolov4 model,it is in a leading position in terms of average detection accuracy,reaching 91.28%,which is higher than the Yolov4 model by 12.99%,the benchmark SSD model by 16.71%,and the Faster-RCNN model by28.75%.