Assessment of Tree Damage with the Forest Health Monitoring (FHM) Method and the Convolutional Neural Network (CNN) Method

So far, the assessment or measurement of tree damage has only been done using the Forest Health Monitoring (FHM) method. This study aims to determine the types of tree damage using Forest Health Monitoring (FHM) and Convolutional Neural Network (CNN) methods. The research was conducted at the TAHURA WAR Utilization Block and the Computer Science Laboratory at FMIPA Lampung University. Measuring the type of tree damage using the FHM method is carried out on trees that are in the FHM cluster. Identification of tree damage types with the CNN algorithm using the MobileNet architecture. The results showed that there were 13 types of tree damage found, with five types of tree damage that were commonly found (> 60 cases): open wounds (218 cases), cancer (94 cases), Broken / Cracks and stems (87 cases), broken or dead branches (73 cases), and loss of dominant shoots (69 cases). As for the identification results with the CNN method, there were nine out of 13 types of damage that obtained precision, recall, and F1 scores of 100%. Thus, five types of dominant tree damage were found, one of which was open wounds (218 cases), and nine types of tree damage obtained high accuracy values.


Introduction
The condition of tree damage is assessed based on the location where the damage was found, the types of tree damage, and the severity level.Damage to a tree is a yardstick to determine whether a tree is healthy or not.Tree damage that occurs over time can result in inhibition of tree growth and development, which also affects forest health [1].The identification of signs and symptoms of tree damage can provide useful information by taking into account the condition of the forest and possible indications of irregularities in area management [2].Damage to trees can be caused by pathogens, pests, and community activities [3].One of the impacts of tree damage is the fall of trees, which can be detrimental both materially and socially [4].Damage to trees should be known as early as possible to allow for treatment or handling of unhealthy trees and minimize tree damage from spreading [5].
So far, the assessment or measurement of tree damage has been carried out using the Forest Health Monitoring (FHM) method.Not least in the TAHURA WAR Utilization Block, the lack of IOP Publishing doi:10.1088/1755-1315/1352/1/012049 2 understanding regarding forest health, especially tree damage, has resulted in a lot of tree damage occurring.The FHM method is a method for monitoring, evaluating, and knowing the status, changes, and trends that occur in a forest ecosystem based on certain measurement indicators [6], [7].This indicates the need for development related to measuring tree damage both manually and digitally.On the other hand, in assessing or measuring tree damage, high accuracy is needed so that the detection results are accurate and precise.To answer these conditions, the Convolutional Neural Network (CNN) method can be used.The CNN method is a computer vision algorithm that can recognize objects, classify images, and perform segmentation.CNN is known as an efficient method and has experienced rapid development in the last few decades in terms of pattern recognition [8].CNN is a trained architecture consisting of several input and output stages.CNN can train and test each incoming image through several processes [9].The purpose of this study was to determine the types of tree damage using the FHM and CNN methods.

Time and Place
Measurement of the types of tree damage was carried out in November 2022 in the Wan Abdul Rachman Forest Park Utilization Block Area (TAHURA WAR), and an assessment of the types of tree damage was carried out at the Computer Science Laboratory, FMIPA, University of Lampung.The tools used include: tally sheet, roll meter, location map, compass, Asus X454W Series laptop with specifications (6.00GB RAM, 500GB hard drive, and AMD E1-6010 processor), Canon EOS 250D camera with CMOS, OPPO mobile camera F9 with dual-camera 16 Megapixels and 2 Megapixels, Windows 7 Ultimate 64-Bit Operating System, and Global Positioning System (GPS).The materials used in this study were tree stands in the TAHURA WAR Utilization Block.

Data Collection
Measurement of the types of tree damage was based on the location where the damage was found, namely on roots, stems, branches, crowns, leaves, shoots, and shoots in the FHM method, carried out on trees in FHM clusters.The research was conducted on 7 cluster plots obtained from calculating the sampling intensity of 0.0025 of the total area of the TAHURA WAR utilization block.If a tree has more than three damages that meet the severity threshold value, the first three damages encountered starting from the root are recorded.The recording of tree damage consists of three sequential codings that describe the location of the damage to the tree, the type of tree damage, and the severity or level of damage to the tree caused.This is performed for a maximum of three damages that meet the severity threshold value, starting with the location with the lowest code.
The dataset is in the form of photos of types of tree damage taken using a Canon EOS 250D camera with CMOS and/or an OPPO F9 mobile camera with dual-camera (16 Megapixels and 2 Megapixels) from several tree samples indicated to be damaged.100 pictures of each type of tree damage.

Data Analysis
The assessment of the types of tree damage in the FHM method was analyzed using the damage index value [10].As for identifying the types of tree damage with the Convolution Neural Network (CNN) algorithm using the MobileNet architecture, The MobileNet architecture is designed for mobile and computer applications using the Tensorflow library.MobileNet is an upgrade from the previous version and uses DSP (Depthwise Separable Convolution) technology [11].The goal is to create a lightweight neural network with reduced parameters.The MobileNet architecture can be seen in Figure 1.

Result and Discussion
Based on the results of measurements and assessments using the FHM method, a number of types of tree damage were found, as shown in Table 1 below.CL = Cluster plot -= No tree damage of this type was found in the cluster Table 1 shows that there are five types of damage to trees that are commonly found (> 60 cases) in the utilization block area, namely: 218 cases of open wound damage, 94 cases of cancer, and 87 cases of broken/cracks and stems (73 cases), and loss of dominant shoots in 69 cases.
The high level of damage that occurs in a forest area is caused by several factors.Factors causing tree damage can be either biotic or abiotic.Biotic factors are factors that cause damage to living organisms, such as pests, diseases, or other organisms that cause damage.Abiotic factors are factors that cause damage from outside living organisms, such as natural disasters, theft, and logging.When these two factors are not a limiting factor in achieving forest sustainability, the health condition of the forest will be maintained [12].Various community activities carried out in the forest have an impact on forest health conditions, especially the level of tree damage [13].Countermeasures against tree damage are needed to prevent the spread of higher damage and minimize the losses experienced.
The CNN method used in this study uses the MobileNet architecture and uses images of 13 types of tree damage, each of which contains 100 images.Image processing of tree damage (preprocessing) that has been taken will then be detected in the model.The model will change the pixel size in the input image according to the format in the model, namely 224 x 224 pixels [14].The change in pixel size aims to expedite the computational process that will run without eliminating the object to be detected or the object to be extracted and used by Epoch 50 with Nvidia Tesla tools [15].Recall, precision, and f1-score values can be seen in Table 2. Table 2 states that all classes (except termite mounds and broken or dead branches) have the best precision value of 100.00%, while these two classes (termite nests and broken or dead branches) have a lower precision value of 96.65%.The higher the precision value, the better the system recognizes it correctly [16].The best recall value is owned by all classes (except broken stems or roots, broken or dead branches, and lianas) at 100%.The two classes (broken stems or roots and broken branches) had a lower recall of 95.65%, followed by the liana class, which had the lowest recall value of 94.74%.The higher the recall value, the more likely it is that the model is successful in detecting the true class [17].
Based on the FHM method, there were five types of dominant damage analyzed by cluster plots: open wounds, cankers, broken stems, broken or dead branches, and loss of dominant shoots.This is consistent with the identification using the CNN method that the five types of tree damage have a high predictive value.Types of open wound damage, cancer, and loss of dominant shoots obtained precision, recall, and f1-scores of 100%.The type of broken stem or root damage obtained a precision value of 100%, a recall of 95.65%, and an f1-score of 97.78%.The type of damage to a broken or dead branch obtains a precision, recall, and f1 score of 95.65%.

Conclusion
Based on the research conducted, it was concluded that there were five types of dominant tree damage found by the Forest Health Monitoring (FHM) method, namely open wounds, cankers, broken/Cracks and stems, broken or dead branches and Loss of dominant shoots, and nine types of tree damage obtained high accuracy values using the Convolutional Neural Network (CNN) method, namely 4 damage to FHM except Broken or dead branches and 4 others namely Conks, fruiting bodies, and other indicators of advanced decay, Resinosis or gummosis, Brooms on roots or bole, Damaged buds, Foliage or shoots, and Discolration of foliage.

Table 1 .
Number of Tree Damage Types Found