Research On Unsupervised Detection Algorithm Of Wood Floor Surface Defects

At present,wood flooring has become one of the important home decoration materials.At the same time,people have higher and higher requirements for the quality of wood flooring.Before leaving the factory,only through the strict defect detection process can the high-quality wood flooring products be sent to the hands of consumers.Manual detection is the earliest and most widely used defect detection method,but the efficiency of manual detection is low and subjective,and the evaluation criteria among workers may not be completely consistent.Therefore,the objective method based on machine vision detection is gradually replacing the manual detection method.Machine vision detection can be divided into traditional detection algorithms and detection algorithms based on deep learning.Because there is no need for prior feature information,the detection algorithm based on deep learning has better applicability,and has gradually become the mainstream method in machine vision detection.The algorithm based on deep learning can be divided into supervised learning and unsupervised learning.Supervised learning needs to use a large number of defect samples for training,but collecting a large number of defect samples in modern industrial production means a lot of economic loss and waste.Unsupervised learning method can only use normal samples to realize defect detection,so unsupervised learning method is more suitable for defect detection in modern industrial production.Therefore,this paper makes an in-depth study on the surface defect detection of wood flooring based on unsupervised learning.The main research work and results are as follows:(1)Using the good image reconstruction ability of the variational autoencoder,an image reconstruction algorithm Rs AE based on residual denoising autoencoder is proposed.In the training phase,the model only uses normal samples,and adds noise to the normal samples as input.The model realizes the reconstruction of input by learning the texture features of normal samples.The floor texture belongs to irregular texture and is very complex,so it is necessary to use the deep model to achieve better reconstruction effect,and the deep model may have the problem of network degradation.Therefore,the residual structure is added to the model to solve this problem,and the position pixel attention module is added to the model to further improve the reconstruction performance of the model.Through the two groups of comparative experiments of real defect reconstruction and image noise reduction,it can be seen that the model proposed in this paper has better image reconstruction effect,and through the ablation experiment,it can be seen that the addition of residual structure and position pixel attention module can improve the reconstruction performance of the model to a certain extent.(2)A multiscale wood floor surface defect detection model MSRs AE based on unsupervised learning is proposed.In the training stage of the model,only the defect-free samples are used and its texture features are learned.When the defect samples are input,the model reconstructs them,and predicts the residual image between the reconstruction result and the input pixel by pixel through the threshold,so as to realize defect detection.In addition,in order to strengthen the detection of small area defects,fuzzy defects and complex lighting conditions,multiscale and cascade structures are added to the model,and the effectiveness of cascade and multiscale structures is verified by ablation experiments.Finally,through the comparison with other models,it can be seen that the model proposed in this paper has lower missed detection rate and false detection rate in the detection results,so the model has better robustness and accuracy.