تشخيص برخي گونه‌هاي درختي از داده‌هاي ابر نقاط ليزر اسكنر زميني با استفاده از الگوريتم‌هاي ماشين بردار پشتيبان و نزديكترين همسايگي

Detection of some Tree Species from Terrestrial Laser Scanner Point Cloud Data Using Support-vector Machine and Nearest Neighborhood Algorithms

شناسايي و تشخيص گونه هاي درختي و اطلاعات مكاني دقيق آنها براي مديريت جنگل‌هاي طبيعي، مصنوعي و همچنين پوشش گياهي شهري امري حياتي و ضروري است. ليزر اسكنر زميني يك سنجنده فعال سنجش از دور مي‌باشد كه پتانسيل توليد اطلاعات مكاني با جزئيات بالا را براي كاربردهايي در زمينه جنگلداري و حفاظت از طبيعت دارد. ليزر اسكنر زميني جزئيات ساختار درختان را در حد زير شاخه توصيف مي‌كند، از اين رو مي‌توان اطلاعات هندسي درختان را با دقت و صحت بالا از داده‌هاي ابر نقاط ليزر اسكنر زميني بدست آورد. روند پيشنهادي در اين مقاله به اين صورت است كه ابتدا از داده‌هاي ابر نقاط ليزر اسكنر زميني سه گونه مختلف درختي كه عبارت است از: Quercus_petraea گونه‌اي از درخت بلوط، Pinus_massoniana گونه‌اي از درخت كاج و Erythrophleum گونه اي از درخت لوبيا، پارامترهاي هندسي اين درختان استخراج شده است. براي هر كدام از اين گونه‌ها 12 داده ابر نقطه ليزر اسكنر زميني موجود بوده است. پس از آنكه پارامترهاي هندسي اين درختان استخراج شدند، با در نظر گرفتن اين پارامترهاي هندسي بعنوان ويژگي و با استفاده از الگوريتم‌هاي ماشين بردار پشتيبان و نزديكترين همسايگي طبقه‌بندي اين سه گونه‌ي درختي انجام شده است. لازم به ذكر است كه ميزان دقت روش هاي استخراج پارامترهاي هندسي درختان توسط داده هاي مرجع كه بصورت غيراتوماتيك توليد شدند، مورد ارزيابي قرار گرفته است. هدف در اين مقاله ارزيابي الگوريتم هاي ماشين بردار پشتيبان و نزديكترين همسايگي با تعداد پارامترهاي هندسي و نمونه هاي آموزشي كمتر براي تشخيص اين سه گونه از هم مي باشد. نتايج ارزيابي دقت طبقه بندي 81% با استفاده از الگوريتم ماشين بردار پشتيبان و دقت 74% با الگوريتم نزديكترين همسايگي را نشان مي‌دهد كه حاكي از قدرت تشخيص نسبتاً خوب روش پيشنهادي براي طبقه‌بندي و تشخيص اين سه گونه مي باشد.

acquisition field reference data using conventional methods due to limited and time-consuming data from a single tree in recent years, to generate reference data for forest studies using terrestrial laser scanner data, aerial laser scanner data, radar and Optics has become commonplace, and complete, accurate 3D data from a single tree or reference trees can be recorded. The detection and identification of tree species and their precise spatial information are essential for the management of natural or man-made forests, and urban vegetation covers. Terrestrial laser scanners are active remote sensing sensors that offer the ability for generating high-level spatial information for forestry and nature conservation applications. A terrestrial laser scanner acquire detailed tree structure even in the sub-branch level. Hence, geometric information of the trees can be obtained with high accuracy from the terrestrial laser scanner point cloud data.The proposed process in this paper is to first use the laser data points of the terrestrial laser scanner of three different tree species: Quercus_petraea oak tree, Pinus_massoniana pine tree and Erythrophleum bean tree. geometric parameters of these trees These include extracted tree height, base canopy height, canopy height, canopy volume and tree diameter profiles. For each species, there were 12 single tree point cloud data of terrestrial laser scanner that were processed by the reference paper provider and the leaves of the trees were considered as noise and deleted. After the geometrical parameters of these trees have been extracted, considering these geometrical parameters (9 geometrical parameters) as a feature and using support vector machine algorithms and nearest neighbor classification of these three tree species done. It is worth noting that the accuracy of the methods for extracting the geometric parameters of trees has been evaluated by reference data that were produced non- automatically. In classification algorithm support vector machine is implemented in MATLAB programming language and RBF kernel is used for separation of three species and from each 12 point clouds of each species 8 point clouds as training data and 4 point clouds as test data are considered. In classifying the nearest neighbor, the value of K is empirically set when the algorithm is most accurate, and same as the SVM method of the 12 clouds available, 8 clouds are considered as training data and the rest of the clouds as test. One of the prominent goals of this study is to investigate the potential of the SVM and KNN for classificaction of tree species using few geometric features and few training samples.The evaluation results indicate the acceptable achieved accuracy 81% for the SVM algorithm and 74% for the KNN algorithm. In both SVM and KNN methods the accuracy of Q. petraea is 100% because the geometrical and structural features of this species are quite different from the other two species, which is clearly visualized in the images and the difference between the two The other class is completely done. The challenge of this classification relates to the other two species because they have almost identical geometrical parameters.The classification results show that the support vector machine algorithm with less training data performs better than the nearest neighbor algorithm in separating these two tree species.