基于改進(jìn)YOLOv8的膠合板單板表面缺陷檢測
打開文本圖片集
關(guān)鍵詞：缺陷檢測；改進(jìn)后的YOLOv8模型（CP-YOLOv8）；CA注意力機(jī)制；CSPPC模塊；WIoUv3；目標(biāo)檢測；輕量化設(shè)計中圖分類號：S776 文獻(xiàn)標(biāo)識碼：A DOI：10.7525/j.issn.1006-8023.2025.04.012
Abstract：Inresponseto thecomplexdiversityofsurface defects inplywood veneersandthe dificulties infeature extraction，as well as the large number of parameters and computational costs of deep learning-based defect detection algorithms，which makes efectiveapplication on devices with lower computing power challenging，a detection model forsurface defects（live knots，dead knots，holes，cracks，and notches）in veneers based onan improvedYOLOv8n is constructed.Toenhance thedetectionaccuracyand lightweightperformanceof the model，improvementsare made to the plywood veneer surface defect detection model.First，anew eficient atention mechanism（coordinate atention，CA）is adopted，which can enhance the acuracyoffeature extraction and the network's spatial information perception ability whileavoiding excesivecomputational burden；secondly，anovel structure basedon partialconvolution（PConv）is proposed- -CSPPC（CSP（coross stage partial） pyramid convolution），it to improve computational efficiency and the fusion capability of multi-scale features； finally，an improved weighted intersection over union loss function- -WIoUv3， it is introduced，which enhances the model'slocalizationaccuracyand robustness.Experimentalresults showthat the improved YOLOv8 model（CP-YOLOv8）performs excellently in the task of detecting surfacedefects in plywood veneers，achieving an average precision mean （mAP）of 93.8% ，an increase of 0.9% over the original model，while reducing the model's floating-point operations （GFLOPs）and parameter count to 7.2Gand2.58 M，respectively，a reductionof O.9 GandO.42 M，which can fully meet practical application needs and provide an eficient，accurate，and lightweight solution for quality inspection of plywood veneers.
0 引言
在全球制造業(yè)加速向自動化、智能化深度轉(zhuǎn)型的時代背景下，木材加工行業(yè)正經(jīng)歷著深刻變革，而膠合板生產(chǎn)作為其中的關(guān)鍵分支，在建筑、家具和裝飾等眾多領(lǐng)域發(fā)揮著不可或缺的作用。（剩余13682字）