Research On Tree Leaf Feature Recognition Based On Additive Algorithm And Gray Co-occurrence Matrix

Forests are the main body for protecting the balance of terrestrial ecosystems.Tree leaves are vital carbon sequestration reservoirs.Effective use and protection of forest resources have become an important issue at the moment.Precise identification of forest species is the key to the construction of ecological forest belts,and it is of great significance to the monitoring of forest resources,the protection of ecological diversity,and the construction of smart forestry.However,because the leaf shape and texture veins are similar and affected by factors such as season and environment,traditional empirical observation methods are difficult to identify tree species.Therefore,integrating machine vision into the field of forestry tree species recognition makes it an important way for automatic classification.Based on the gray level co-occurrence matrix,this study constructs a tree species recognition model based on BP neural network,and it is verified by experiments that tree species identification can be accurately and effectively realized.Collect the leaves of 10 broad-leaved tree species in the forest area at the experimental forest farm of Northeast Forestry University as the research object,including(Diamond Populus,Lonicera Lonicerae,Salix,Astragalus,Ulmus pumila,Maple,Aster,White Birch,Maple and Mongolian Oak).Through an independently designed and developed image acquisition hardware system,it is used to obtain two-dimensional digital images of leaf samples,and to perform grayscale,binarization,and edge detection processing.At the same time,a new addition method was proposed to segment the original grayscale image of the leaf from the background area of the entire image.This method retained the complete feature information of the grayscale image of the leaf,and the leaf area was completely separated from the background.It provides solid theoretical support for subsequent research on image feature extraction and tree species recognition.The addition method can also be applied to other image segmentation fields.For the pre-processed 2D tree leaf images,the geometric and texture feature information is extracted.The gray-scale co-occurrence matrix method is used to obtain the leaf texture feature values from multiple directions and angles.Because the leaf texture is delicate,considering the program running time,a d=1;g=64;?=0°,45°,90°,135°gray-scale co-occurrence matrix.There are seven terms in texture feature descriptor,which can be expressed as:entropy,significant clustering,correlation,deficit moment,moment of inertia,angular second moment and clustering shadow.There are six geometric descriptors,which can be expressed as:aspect ratio,eccentricity,density,shape parameters,rectangularity,and sphericity.A three-layer BP neural network model was constructed,and the normalized 13 eigenvalues were used as network inputs to identify tree species.The 12 normalized eigenvalues are used as the network input to identify the tree species,When the hidden layer is determined by the trial and error method,the recognition rate is 91.25%,and the optimal solution is obtained.The experimental results show that the tree species recognition system constructed in this study is faster,more accurate,and more efficient than the traditional artificial feature recognition method to automatically classify 10 common broad-leaved tree species in Northeast China,providing support for the automatic identification of forestry tree species images.