A Study On Algorithm Of Log Detection

Based on the current situation of fragile ecological environment and the scarcity of forest resources,it is necessary to improve the utilization of wood,use the limited forest resources effectively,to alleviate the increasing tension of wood supply and demand contradictions.It's impressively significant in timber rational utilization and saves national forest resources by quickly and effectively log defects detection and log classification.An algorithm to detect log defects nondestructively,quickly and efficiently was researched in this thesis,the main points and research work are as follows.Based on the research and analysis of common image features such as point features,line features,area features,color features,and texture features,extract log defects features by SIFT(Scale Invariant Feature Transform)algorithm.SIFT features which belong to the point features,have advantages such as invariant to image scale,rotation,and illumination.Thus,image features can be saved commendably.Simulated experiments and analysis were performed through actual log defect images.The BOW(Bag of Word)model which was originally applied in document categorization area was introduced into this study.Based on the BOW model,a log defects automatic classification algorithm was proposed.Firstly,clustering the SIFT features which were extracted from log defects through K-means clustering algorithm,get the visual dictionary of log images.Then,according to this dictionary,count the TF-IDF(Term Frequency-Inverse Document Frequency)for each image,in order to build a BOW model of log images.Thus,each log image can be represented as a vector.Performing automatic classification experiment for log defects by SVM(Support Vector Machines)classifier.Firstly,training the SVM classifier by log defect images which in the training set and has been calibrated.Then,automatically classify the images in the test set by the trained SVM classifier,and the results obtained were statistically analyzed.Three representative defects,dead knot,live knot and wormhole were chosen as the sample for simulation.Experimental results show that algorithm in this thesis can auto classify these three defects by 91% accuracy stably,which can meet the demands of quickly,nondestructively log defects detection.