Adaptive Ant Colony Optimization on Mango Classification Using K-Nearest Neighbor and Support Vector Machine

— Leaves recognition can use an image edge detection method. In this research, the classification of mango gadung and manalagi will be performed. In the preprocess stage edge detection method using adaptive ant colony optimization method. The use of adaptive ant colony optimization method aims to optimize the process of edge detection of a mango leaves the bone image. The application of ant colony optimization method on mango leaves classification has successfully optimized the result of edge detection of a mango leaves the bone structure. Results showed edge detection using adaptive ant colony optimization method better than Roberts and Sobel method. The result an experiment of mango leaves classification with k-nearest neighbor method get accuracy value equal to 66,25%, whereas with the method of support vector machine obtained accuracy value equal to 68,75%.


I. INTRODUCTION
Plants are the most important part of life on earth.Plants are used to provide oxygen for breathing, as food, fuel, medicine, cosmetics, and more.The plants also help in regulating the climate, providing habitats for animals, preventing a flood, as food for insects and other animals.One type of plant that is widely used is a fruit plant.This research was conducted to discuss the classification of mango species, especially gadung and manalagi mango.The process of mango classification is done by identifying based on the image of mango plant leaves.The leaves are one part of the plant that is often used to classify the types of plants (Jabal, Mohamad, Suhardi, Salehuddin, & Illiasak, 2013).In this way, the leaf pattern recognition steps can be recognized by recognizing the structural characteristics of the leaves such as the shape and texture of the bone leaf structure.Each type of plant has different leaves that can be used as a reference to classify each type of plant (Fu & Chi, 2006).
Edge detection is the process of finding information from the edge of the image that is the basic step in most image processing applications (Gonzales & Wood, 1992).The edges of the image can be considered as the boundary between two different areas.Many approaches have been used to investigate the edges of images.Some commonly used edge detection methods are Roberts, Sobel, Prewitt and canny (Charu & Sunanda, 2013) (Om, Hanmandlu, Sultania, & Dhruv, 2010).Ant colony optimization is an optimization method that can be used to detect image edges.This method is a method that mimics the heuristic behavior of ants to solve discrete optimization problems (Dorigo, Birattari, & Stutzle, 2006).As Rahebi et al, research by combining optimization of ant colonies and genetic algorithms in improving ant dissemination to improve convergence (Rahebi, Elmi, & Shayan, 2010).
In the previous research, adaptive ant colony optimization is done by dividing the image into 4 areas and ants are divided based on the gradient value of each region but the initial an ant dissemination is still done randomly, precision and recall of adaptive ant colony optimization reach 76.98 % and 96.8 % (Liantoni, Kartika, & Tri, 2014).In 2015, research was conducted for ant colony optimization modification with initial an ant dissemination based on the gradient value of image pixels, this research obtained an average value of peak signal to noise ratio (PSNR) of 12.724 (Liantoni, 2015).While research on the classification of plant species has also been done.In 2017, the researchers themselves have done the grouping of herbal leaves using hierarchical clustering method based on the moment invariant feature, the use of hierarchical clustering method obtained clustering accuracy value of 87%, whereas by using K-Means method obtained clustering accuracy of 81% (Liantoni & Cahyani, 2017).In 2015 researchers conducted research on the classification of herbal leaves Bayes Classifier method and K-This research compares the two methods to calculate the accuracy of both methods Naïve Bayes Classifier method obtained accuracy of 75% while using the Neighbor method obtained an accuracy of 70.83%.(Liantoni & Nugroho 2015) researchers also conducted research on the classification of the leaves with the improveme of image features using the K-Nearest Neighbor testing the result of leaves classification from image which is on dataset has been built to get accuracy value about 86,67% (Liantoni 2015, Suastika et al classified mango plants based on leaf bone using Local Binary Pattern method (LBP), the highest accuracy is reached (Riska, Cahyani, & Rosadi, 2015) In this research, edge detection process that generally uses conventional method will be changed using ant colony optimization method.The method proposed in the process of optimizing the deployment of ants.In general ant colony optimization method, the initial ant positions are randomly distributed, this triggers an imbalance in the distribution of ants in the path ant colony optimization method is proposed to optimize the result of edge detection of mango bone leaves structure so that it can be obtained better classification of mango plant species.In this research K-Nearest Neighbor and Machine method is used for process.

A. Adaptive Ant Colony Optimization
In this research, the detection process of mango leaf edge is done by using adaptive ant colony Journal of Information Systems Engineering and Business Intelligence 76 hers conducted research on ication of herbal leaves Using Naïve -Nearest Neighbor.compares the two methods to calculate the accuracy of both methods, the use of hod obtained an while using the K-Nearest method obtained an accuracy of 70.83%.(Liantoni & Nugroho 2015).In addition, researchers also conducted research on the classification of the leaves with the improvement Nearest Neighbor, testing the result of leaves classification from the image which is on dataset has been built to get (Liantoni, 2015).In 2015, Suastika et al classified mango plants based cal Binary Pattern method he highest accuracy is reached 78.5%.(Riska, Cahyani, & Rosadi, 2015).
In this research, edge detection process that conventional method will be changed using ant colony optimization method.The method proposed in the process of optimizing oyment of ants.In general ant colony optimization method, the initial ant positions are randomly distributed, this triggers an imbalance in istribution of ants in the pathfinding.Adaptive ant colony optimization method is proposed to sult of edge detection of mango bone leaves structure so that it can be obtained a lant species.In this eighbor and Support Vector method is used for the classification Flow Mechanism of Adaptive ACO METHODOLOGY Adaptive Ant Colony Optimization the detection process of mango leaf edge is done by using adaptive ant colony optimization method.This method refers to research that has been done earlier in 20 method is done by making changes to traditional ant colony optimization methods.The proposed method by dividing the image into 4 areas is equal then the ant will be spread by following the number of edge potentials of each area based on the value of the gradient.Because the edge is assumed to be at a high gradient value Sunanda, 2013).This method is expected to optimize the results of edge detection on mango leaf bone.Figure 1 is an adaptive ant colony optimization diagram.
In this research, instead of using Ant System, we prefer to use ACS algorithm.the pseudorandom proportional rule to optimize ant's movement.The pseudoran rule uses user-defined threshold ( q between 0 to 1, to complement traditional random proportional rule.On each ant's movement, it is needed to randomly generate q which value is distributed between 0 and 1.If q is greater, then random proportional rule is us movement.However, if q < q 0 , according to transition that maximizes random proportional rule is given in equation(1).
The probability of ant movement to row i and column j (Pij) equals to of pheromone ( ) and heuristic information ( which is divided by total multiplication of its 8 connectivity neighborhood.Both pheromo heuristic information is equipped with pheromone weighting factor (α) and heuristic information weighting factor (β).In addition, ACS has two pheromone update mechanisms, which are local pheromone update and global pheromone update.Local pheromone update is firstly introduced in ACS and considered as the most interesting contribution of ACS (Jiang, Weiyu, & Shengli, 2008).Formulas for local pheromone and global pheromone updates are defined in equation ( 2) and equation (3)., = (1 − ) , + , = (1 − ) , + ∆ , Local pheromone update aims to diversify the search performed by subsequent ants during an iteration (Dorigo et al. 2006).It is using pheromone decay (φ) and pheromone init ( ) to decrease the pheromone concentration on the traversed edges.Therefore subsequent ants can produce different solutions.Whereas pheromone evaporation rate ( ) and total deposit pheromo optimization method.This method refers to research that has been done earlier in 2015.The method is done by making changes to traditional ant colony optimization methods.The proposed method by dividing the image into 4 areas is equal then the ant will be spread by following the number of edge potentials of each area based on of the gradient.Because the edge is assumed to be at a high gradient value (Charu & .This method is expected to optimize the results of edge detection on mango adaptive ant colony , instead of using Ant System, we prefer to use ACS algorithm.Since ACS has pseudorandom proportional rule to optimize seudorandom proportional shold ( q 0 ), that value is between 0 to 1, to complement traditional random proportional rule.On each ant's movement, it is needed to randomly generate q which value is distributed between 0 and 1.If q is greater, then the random proportional rule is used to decide ant's , an ant should move according to transition that maximizes , , .The andom proportional rule is given in equation( 1).

∈ Ω
(1) The probability of ant movement to a pixel on ) equals to a multiplication ) and heuristic information (h ) which is divided by total multiplication of its 8 connectivity neighborhood.Both pheromone and equipped with pheromone weighting factor (α) and heuristic information weighting factor (β).In addition, ACS has two pheromone update mechanisms, which are local pheromone update and global pheromone update.
update is firstly introduced in sidered as the most interesting (Jiang, Weiyu, & Shengli, .Formulas for local pheromone and global pheromone updates are defined in equation ( 2) and (2) (3) Local pheromone update aims to diversify the search performed by subsequent ants during an .It is implemented by using pheromone decay (φ) and pheromone init ) to decrease the pheromone concentration on the traversed edges.Therefore subsequent ants can produce different solutions.Whereas pheromone ) and total deposit pheromone (∆ , ) are used to implement global pheromone update.

B. Classifier
The last stages of this research of classification of herbal leaf image pattern.The K-Nearest Neighbor method and the Vector Machine are used for the classification process.KNN algorithm determines the distance value in testing data testing with training data based on the smallest value of the nearest neighboring value (Goujon, Chaoqun, & Jianhong, 2007).According to Prasetyo, SVM method (Support Vector Machine) is a theory of statistical learning and can give better results than other methods SVM can work well on data with a high dimensional (Prasetyo & Eko, 2012) of both methods will be compared to determine the best accuracy of the method used. in the leaves classification process are shown in Figure 2.

B. Adaptive Ant Colony Optimization Result
In this research, the adaptive optimization method was used as a substitute for conventional edge detection method.adaptive ant colony optimization used in accordance with research that has been done (Liantoni, Kartika, & Tri, 2014)

Figure. 2 .
Figure. 2. The leaves classification process III.RESULT Tests conducted have a purpose to deter the level of system success in the mango leaves.Scenario testing is done with the amount of mango leaves image data of 400 consisting of 200 leaves of gadung mango and 200