| As an integral part of forestry ecosystem,trees play a vital role in improving people’s living environment,providing social services and material production.Traditional tree detection methods require large equipment,are not easy to carry and are subject to large environmental factors.For trees beyond the level of a single tree,experienced observers are still needed,and professional requirements are high.When climate,terrain and other factors change,the accuracy and applicability of tree detection work also need to be further improved.In this paper,the overall image data set of trees is constructed by means of independent photography and online acquisition.Considering the differences in the morphological performance of trees at different growth stages,the images of trees at mature stages are taken as the main acquisition objects in image collection.In addition,data enhancement processing was carried out by means of diagonal inversion,affine transformation,color adjustment,local cropping,and fuzzy processing.A total of 15 types of trees were selected for detection to achieve the acquisition of specific location distribution information of trees.Based on the whole data set of trees,this paper proposes a tree detection method based on Faster R-CNN,which can solve the problem of poor detection effect caused by different illumination conditions and tree shape occlusion.By comparing the detection effect of different data sets on Faster R-CNN.It can be found that after the training set of the original tree data set is processed by the data-enhanced operation,the detection accuracy of the test set has been improved to a certain extent.The MAP of the network model trained on this training set reaches 90.28%.The combination of the training set and the test set is used as a common tree image data set to optimize the detection model.In addition,the detection accuracy of the test set after light treatment and occlusion treatment reached 93.91% and 93.94%,and the accuracy of tree detection improved to 95.09% under the action of light treatment and occlusion treatment.Compared with other models(SSD,YOLOV3),it has a better detection effect,which is conducive to the advancement of the extraction of tree-related attribute characteristic information in the later stage,reducing the process of manual intervention,and liberating manpower to a certain extent.At the same time,on the basis of this paper trees detection,through the trees the overall image of the form,makes an analysis of the health,the most common location of tilt,integrity,and color as evaluation indexes,through the specific judgment and analysis of these three factors,to evaluate trees form condition,obtain the specific problems of the trees,To achieve the trees in the initial damaged state can be timely tracked to promote the later maintenance work,for the subsequent growth environment of trees,underground soil conditions,internal conditions of trees and other aspects of the analysis to pave the way. |
