VOL. 12, NO. 8, APRIL 2017 ISSN 1819-6608 ARPN Journal of Engineering and Applied Sciences © 2006-2017 Asian Research Publishing Network (ARPN). All rights reserved. www.arpnjournals.com 2730 DETECTION OF MULTIPLE MANGOES USING HISTOGRAM OF ORIENTED GRADIENT TECHNIQUE IN AERIAL MONITORING Nursabillilah Mohd Ali 1,2 , Mohd Safirin Karis 1,2 , Nur Maisarah Mohd Sobran 1,2 , Mohd Bazli Bahar 1,2 , Oh Kok Ken 1,2 , Masrullizam Mat Ibrahim 3 and Nurul Fatiha Johan 1,2 1 Faculty of Electrical Engineering, Universiti Teknikal Malaysia Melaka, Durian Tunggal, Melaka, Malaysia 2 Robotics and Industrial Application Automation Research Laboratory, Durian Tunggal, Melaka, Malaysia 3 Faculty of Electronics and Computer Engineering, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia E-Mail: nursabilillah@utem.edu.my ABSTRACT The project uses shape identification algorithm and Histogram of Oriented Gradient principle to detect and count the total number of mango on its tree using a quad copter with an attachable webcam. The traditional method in harvesting mango has its limitation which leads to the degradation of harvested mango’s quality. As a result, the rate of production and the structure of the tree will be dampening. Hence, usage of image processing algorithm could be a solution for a better and more precise mango’s pre-harvesting process. It differentiates the mango and its leaf based on the images captured on real scene and thus forecast the growth rate of the mango tree for time being. Tallness of the mango tree and location of mango would not affect farmer’s capability to inspect the mango as the drone hovers according to user’s intention. It is expected to provide an alternate review for the mango grower, agricultural developer and investor. Keywords: harvesting, histogram of oriented gradient, image processing, shape. INTRODUCTION Harvesting is one of the main processes in agriculture sector. In harvesting a mango, ripeness of the fruit will determine the output taste either sours, sweet, creamy or mild bitterness. The normal inspection of mangoes ripeness usually done based on manual inspection by the farmer. However, the height of a mango tree, abundant of branches and leaves might hindrance the inspection process. With this motivation, we apply the Histogram of Oriented Gradient method in detecting multiple mangoes at the tree branches and implementing the quad copter as a mechanism in reaching the height of the tree. Researches in harvesting mango in agriculture cover wide area. Major researches are in detecting the level of ripeness of the mango in [1, 2]. After that, issue in grading or classifying the mangos according to the detected ripeness as mention in [3] and [4]. Other than that, problems arise in plucking mango in [5] and peeling numbers of mango in [6]. There is also large scale of research done in [7] for monitoring the green house plant for Harum Manis Mango. This project focus more on detecting mangoes at branches for monitoring process before the harvesting were done. The idea is to provide early observation to the person in charge before decide whether or not to pluck the mango, while covering the restricted sight that the person in charge had due to mango trees height. In [8] detect mango in the presence of leaves and branches using Randomized Hough Transform and Backpropagation Neural Network, the same method applied by [9]. In [10] implementing the Euclidean distance based classifier in detecting apples at the branch, whereas in [11] used centroid based detection in recognizing orange at the tree. This project proposed histogram of oriented gradient (HoG) based mango detection as algorithm in finding mangoes in the captured images. DETECTION USING HISTOGRAM OF ORIENTED GRADIENT (HoG) Histogram of Oriented Gradient descriptor is actually a feature descriptor widely used in machine vision field for the purpose of detection of object. It calculates the number of occurrences of gradient orientation in localized parts of an image. The main point of HoG is the distribution of intensity gradient and edge directions. It could be done by dividing the image into small connected regions, each compiled with histogram of gradient direction and edge orientations for the pixels involved. Hence, the histograms merged to become a descriptor. The performance could also be improved by contrast- normalizing the local histogram by computing a measure of intensity across a region within the image (a block). The block would normalize the smaller connected region cells within itself. Shadowing or imperfect illumination would probably be reduced by the normalization. The next descriptions of HoF technique are based on [12, 13]. In order to obtain the feature descriptor, the first step taken is filtering the grayscale image I with the following kernels as in Equation. (1) and (2).   1 0 1   x D (1)             1 0 1 y D (2)