Study of Neural Network Training Algorithms in Detection of Wood Surface Defects

Abstract Accurate detection of defects through machine vision improves the economical growth of wood industry. In this paper six common defects on wood surface are considered for study. The quality of the wood images is enhanced by Histogram Equalization method. The contrast enhanced images are subject to Thresholding segmentation which examines the objects in the image and identifies the defect. The segmented images are cropped in to small blocks. SFTA feature extraction method is accomplished to extract 21 texture features from the wood images. The extracted features are fed in to the training algorithms such as Levenberg-Marquardt, Scaled Conjugate Gradient, Gradient Descent with Adaptive Learning Rate, Bayesian Regularization and Resilent Backpropagation. The performance of the training algorithms is analyzed with several performance metrics. The result obtained shows a considerable improvement in accuracy of 98.2 % by Bayesian Regularization tool.