ﻣﺮوري ﺑﺮ ﺧﻮارزﻣﯽ ﻫﺎي ﺗﺸﺨﯿﺺ و ﻣﺮزﺑﻨﺪي ﺗﺎج درﺧﺘﺎن ﺑﺮ روي ﺗﺼﺎوﯾﺮ ﻫﻮاﯾﯽ

A review of algorithms for trees crown detection and delineation on aerial images

ﺳﺎﺑﻘﻪ و ﻫﺪف: ﻣﺪﯾﺮﯾﺖ ﮐﺎرآﻣﺪ ﺟﻨﮕﻞ ﻧﯿﺎز ﺑﻪ اﻃﻼﻋﺎت دﻗﯿﻖ و ﺑﻪ ﻣﻮﻗﻊ دارد. ﺗﺼﺎوﯾﺮ ﺳﻨﺠﺶ ازدوري ﺑﺎﻗﺪرت ﺗﻔﮑﯿﮏ ﻣﮑﺎﻧﯽ ﺑﺎﻻ )ﻋﮑﺲ ﻫﺎي ﻫﻮاﯾﯽ(، ﭘﺘﺎﻧﺴﯿﻞ ﺑﺎﻻﯾﯽ در ﺑﺮآورد اﻃﻼﻋﺎت ﺟﻨﮕﻞ ﺑﺎ ﻫﺰﯾﻨﻪ ﮐﻢﺗﺮ را دارﻧﺪ. ﻧﮑﺘﻪ ﻣﻬﻢ در اﯾﻦ زﻣﯿﻨﻪ، اﺟﺮاي روش ﺑﻬﯿﻨﻪ ﻃﺒﻘﻪﺑﻨﺪي و اﺳﺘﺨﺮاج اﻃﻼﻋﺎت ﻣﻔﯿﺪ از اﯾﻦ دادهﻫﺎي ﺑﺎ ﺗﻮان ﺗﻔﮑﯿﮏ ﻣﮑﺎﻧﯽ زﯾﺎد اﺳﺖ. روش ﻫﺎي ﻃﺒﻘﻪﺑﻨﺪي ﺷﯽ ﭘﺎﯾﻪ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ اﺳﺘﺨﺮاج و ﺟﺪاﺳﺎزي ﭘﺪﯾﺪهﻫﺎي ﻫﻤﮕﻦ و ﻃﺒﻘﻪﺑﻨﺪي آن ﻫﺎ، داراي ﻧﺘﺎﯾﺞ ﺑﻬﺘﺮي ﻧﺴﺒﺖ ﺑﻪ ﻃﺒﻘﻪﺑﻨﺪي ﭘﯿﮑﺴﻞ ﭘﺎﯾﻪ اﺳﺖ. ﺑﺎ ﻗﻄﻌﻪ ﺑﻨﺪي و ﻃﺒﻘﻪﺑﻨﺪي ﺗﺎج درﺧﺘﺎن در ﺗﺼﺎوﯾﺮ ﺑﺎﻗﺪرت ﺗﻔﮑﯿﮏ ﺑﺎﻻ اﻃﻼﻋﺎت ﺑﺎارزﺷﯽ از ﺟﻨﮕﻞ اﺳﺘﺨﺮاج ﻣﯽﺷﻮد ﮐﻪ ﻣﯽ ﺗﻮاﻧﺪ در ﻣﺪﯾﺮﯾﺖ ﺟﻨﮕﻞﻫﺎ ﻣﻔﯿﺪ واﻗﻊ ﺷﻮد. ﻣﻮاد و روش ﻫﺎ: اﯾﻦ ﭘﮋوﻫﺶ از ﻧﻮع ﻣﺮوري ﺑﻮده ﮐﻪ ﺑﻪ روش ﮐﺘﺎﺑﺨﺎﻧﻪ اي و ﺑﺎ اﺳﺘﻔﺎده از ﻣﺪارك و ﺳﻮاﺑﻖ ﻣﻮﺟﻮد، ﺧﻮارزﻣﯽ ﻫﺎي ﺗﺸﺨﯿﺺ و ﻣﺮزﺑﻨﺪي ﺗﺎج درﺧﺘﺎن ﻣﻮرد ﺑﺮرﺳﯽ ﻗﺮار ﮔﺮﻓﺘﻪ اﺳﺖ. ﯾﺎﻓﺘﻪ ﻫﺎ: در اﯾﻦ ﭘﮋوﻫﺶ ﺧﻮارزﻣﯽ ﻫﺎي ﻣﺨﺘﻠﻔﯽ ﺑﺮاي ﻗﻄﻌﻪﺑﻨﺪي ﺗﺎج درﺧﺘﺎن اراﺋﻪ ﺷﺪه و ﺑﺎ ﯾﮑﺪﯾﮕﺮ ﺑﺮاي ﺗﻌﯿﯿﻦ ﻗﻮت و ﺿﻌﻒ ﻫﺮ ﮐﺪام ﻣﻘﺎﯾﺴﻪ اي ﺻﻮرت ﮔﺮﻓﺘﻪ اﺳﺖ. ﺧﻮارزﻣﯽﻫﺎي ﻗﻄﻌﻪﺑﻨﺪي در دو ﻣﺮﺣﻠﻪ ﺗﺸﺨﯿﺺ و ﻣﺮزﺑﻨﺪي ﺗﺎج درﺧﺘﺎن اﻧﺠﺎم ﻣﯽ ﺷﻮد. ﻫﻢ ﭼﻨﯿﻦ ﮐﺎراﯾﯽ و ﺷﺮاﯾﻂ ﻫﺮ ﮐﺪام از ﺧﻮارزﻣﯽ ﻫﺎ ﻣﻮرد ﺑﺮرﺳﯽ و ﺑﺤﺚ ﻗﺮار ﮔﺮﻓﺘﻪ اﺳﺖ. ﻫﺮ ﮐﺪام از ﺧﻮارزﻣﯽ ﻫﺎي ﺗﺸﺨﯿﺺ )ﻓﯿﻠﺘﺮ ﺣﺪاﮐﺜﺮ ﻣﺤﻠﯽ، دوﺗﺎﯾﯽ ﺳﺎزي ﺗﺼﻮﯾﺮ، ﺗﺤﻠﯿﻞ ﻣﻘﯿﺎس، ﺗﻄﺒﯿﻖ اﻟﮕﻮ( و ﻣﺮزﺑﻨﺪي اﻟﮕﻮي دره، ﻗﻄﻌﻪ ﺑﻨﺪي ﺣﻮﺿﻪ آﺑﺨﯿﺰ، روﯾﺶ ﻧﺎﺣﯿﻪ اي( داراي ﻣﻌﺎﯾﺐ و ﻣﺰاﯾﺎﯾﯽ ﻫﺴﺘﻨﺪ. ﻧﺘﯿﺠﻪ ﮔﯿﺮي: در اﯾﻦ ﻣﻄﺎﻟﻌﻪ ﻣﺸﺨﺺ ﺷﺪ ﮐﻪ از ﺑﯿﻦ ﺧﻮارزﻣﯽ ﻫﺎي ﺗﺸﺨﯿﺺ، ﺗﻄﺒﯿﻖ اﻟﮕﻮ ﺑﺮاي اﯾﻦ ﮐﻪ ﻣﺘﻨﺎﺳﺐ ﺑﺮ ﻫﺮ ﻧﻮع ﺗﺎج درﺧﺘﯽ اﺳﺖ ﻣﯽ ﺗﻮاﻧﺪ در ﺟﻨﮕﻞ ﻫﺎي ﺷﻤﺎل ﻣﻮرداﺳﺘﻔﺎده ﻗﺮار ﮔﯿﺮد و اﮔﺮ ﻣﻨﻄﻘﻪ اي ﺟﻨﮕﻞ ﮐﺎري ﺷﺪه ﺑﺎﺷﺪ ﻓﯿﻠﺘﺮ ﺣﺪاﮐﺜﺮ ﻣﺤﻠﯽ ﺑﻬﺘﺮﯾﻦ ﮐﺎرﺑﺮد را دارد. از ﺑﯿﻦ ﺧﻮارزﻣﯽ ﻫﺎي ﻣﺮزﺑﻨﺪي، ﻗﻄﻌﻪ ﺑﻨﺪي ﺣﻮﺿﻪ آﺑﺨﯿﺰ و اﻟﮕﻮي دره ﺑﺮاي ﺟﻨﮕﻞ ﻫﺎي زاﮔﺮس ﻣﻨﺎﺳﺐ اﺳﺖ و روﯾﺶ ﻧﺎﺣﯿﻪ اي ﻣﻨﺎﺳﺐ ﺟﻨﮕﻞﻫﺎي ﻣﺘﺮاﮐﻢ اﺳﺖ ﺑﻪ ﺷﺮﻃﯽ ﮐﻪ ﻓﻀﺎي ﺑﯿﻦ ﺗﺎج ﻫﺎ دﯾﺪه ﻧﺸﻮد

Abstract1 Background and Objectives: Efficient forest management needs accurate and timely information. Remote sensing images with high spatial resolution have high (aerial images) potential for estimating forest data with lower cost. The important point in this context is performance of optimal method for classification and useful information extraction from the high spatial resolution data. Object-based classification methods by extraction and separation of homogeneous phenomena and their classification have better results compared to pixel-based classification. By segmentation and classification of individual tree crowns in high spatial resolution aerial images valuable information from forest can be obtained which can be useful in forest management. Materials and Methods: This study is a review article which has been done by library method and using existing documents, algorithms for trees crown detection and delineation. Results: In this study different algorithms were presented for segmenting individual tree which were compared to one another to determine the strengths and weaknesses of each one. Segmentation algorithms are divided into two steps: tree-crown detection and delineation. Also efficiency and condition of each algorithm is discussed. Each detection algorithms (Local Maximum Filtering, Image binarization, Template Matching, and Scale Analysis) and delineation algorithms (Valley-Following, Watershed Segmentation and Region-Growing) has disadvantages and advantages. Conclusion: This study specified that among the detection algorithms, Template Matching can be used in north forests of Iran because suits every type of tree crown. But in plantation areas the local maximum filtering algorithm is the best algorithm. Of delineation algorithms, Watershed-Segmentation and Valley Following are suitable for Zagro