Teguh Bharata Adji

UAV obstacle avoidance using potential field under dynamic environment

In this paper, the potential field principle is applied to several UAVs (Unmanned Aerial Vehicles) for optimal path planning. Each UAV has its own goals and it is used the attractive potential field to reach the goals. On the contrary, each UAV is considered as obstacle for other UAVs that must be avoided. In this research, there are two types of obstacles, i.e the static and dynamic. The repulsive potential field principle is used to avoid for both static and dynamic obstacles. The whole method is implemented into Parrot AR Drone 2.0 Quadcopter model of UAV and simulated in Gazebo Simulator by Robot Operating System (ROS). The results of this research are to control the thrust of quadcopter so that it is in flying position, the value of value of roll (α) or pitch (β) must set to not equal or approaching 90° or -90° and not between -180° and 180° or between -90° and 90°. Based on the dynamic obstacle performance test with parameter tuning, the optimal avoidance is when the η value is 7.8 while when in static and dynamic test with parameter tuning, the optimal avoidance is when the η value is 7.9 noted by the fastest time and the shortest path.

Optic Disc Segmentation Based on Red Channel Retinal Fundus Images

Glaucoma is a one of the serious diseases that occurs in retina. Early detection of glaucoma can prevent patients from blindness. One of the techniques to support the diagnosis of glaucoma is developed through the detection and segmentation of optic disc area. Optic disc area is also useful in assisting automated detection of abnormalities in the case of diabetic retinopathy. In this work, extracted red channel of colour retinal fundus images is used. Median filter is used to reduce noises in the red channel image. Segmentation of optic disc is conducted based on morphological operation. DRISHTI-GS dataset is used in this research works. Results indicate that the proposed method can achieve an accuracy of 94.546% in segmenting the optic disc.

Multiple layer data hiding scheme based on difference expansion of quad

Performance of the difference expansion of quad scheme can be improved by applying IRDE.Multiple-layer embedding can be done by applying scheme of reversible data hiding.The combined of scheme difference expansion of quad with IRDE can be to improve the performance in multiple-layer embedding.The performance of the combined scheme difference expansion of quad with IRDE has a better than the combined of difference expansion of quad with RDE in single-layer or multiple-layer embedding. For the past few years, the schemes of data hiding are growing rapidly. Generally, data hiding performs well on common images but it does not provide satisfying results on distortion sensitive images such as medical, military, or forensic images. This is because embedding data into an image can cause permanent distortion after extraction (irreversible). As a solution, a certain scheme is required for the process of embedding data into an image, such as reversible data hiding (RDH). One well-known RDH scheme is difference expansion, a simple and easy to implement scheme. In this study, a new scheme, multiple-layer embedding based on difference expansion of quad, is proposed, focusing on increasing capacity and visual quality of data hiding by reducing difference value in pixel with improved reduced difference expansion (IRDE). The proposed scheme has been evaluated with 14 grayscale images, consisting of six common images and eight medical images. Results show that the proposed scheme has higher capacity and better visual quality compared to the original scheme and previously similar schemes.

Segmentation of exudates based on high pass filtering in retinal fundus images

The World Diabetes Foundation has predicted that more than 439 million people in 2030 will suffer from diabetes. Long-term diabetics can lead to the damage of retinal blood vessels, known as diabetic retinopathy, the leading cause of blindness in developing countries. One of the clinical features of diabetic retinopathy is exudate. Exudates have similar characteristic with optic disc. Therefore, in this research work, removal of optic disc is conducted to reduce false positive of exudates detection. The optic disc detection is done by finding the small area of the optic disc which is enlarged to obtain its total area. Green channel that contains useful information for exudates detection is filtered based on high pass filter. Afterwards, segmentation of exudates is conducted by using thresholding and morphological operations. Final result of exudates is validated with ground truth images by measuring accuracy, sensitivity and specificity. The results show that proposed approach for exudates detection achieves accuracy, sensitivity and specificity of 99.99%, 90.15% and 99.99%, respectively. This result indicates that the proposed method successfully detects exudates and is useful to assist the ophthalmologists in analysing retinal fundus image especially for exudates detection to diagnose diabetic retinopathy.

Dynamic uav path planning for moving target intercept in 3D

Dynamic path planning is one of the challenging research problems which is needed to guide UAV (Unmanned Aerial Vehicle) for moving target intercept. This research is to develop an algorithm for moving target intercept with obstacle avoidance in 3D. The algorithm which is called L+Dumo Algorithm integrate a modified Dubins Algorithm and Linear Algorithm. The simulation is conducted using one UAV and one moving target with an obstacle of cylindrical shape in between both objects. The result shows that the algorithm can guide UAV to approach the moving target with the accuracy of 67.417% L+Dumo algorithm will solve dynamic path planning problem for moving target intercept in 3D with minimum complexity of computation.

Adult image classifiers based on face detection using Viola-Jones method

This research consists of a classification system to determine the adult and benign image. Adult image in this research was defined as image that is perceived as pornographic by Indonesian people. The method in this research combines face detection and HS skin color detection on an image. Face detection is done by using the Viola-Jones method. After face detection process, skin detection is performed on the image. Based on the results of face and skin detections, a set of features is extracted and inserted into the classifier. The classification used in determining adult or benign image will be based on the percentage of face area in the image, the position of face in the image, and the percentage of the skin color in the image. For each feature, the threshold value is defined in this research. The results of the classifier is whether an input image is benign or adult images. From 30 sample images, the classification process classifies 5 pieces of images as benign images and 25 images as adult images. False positive rate are 2 images and false negative rate is 1 image with the accuracy of 90%.

Deep residual coalesced convolutional network for efficient semantic road segmentation

This paper proposes a deep learning-based efficient and compact solution for road scene segmentation problem, named deep residual coalesced convolutional network (RCC-Net). Initially, the RCC-Net performs dimensionality reduction to compress and extract relevant features, from which it is subsequently delivered to the encoder. The encoder adopts the residual network style for efficient model size. In the core of each residual network, three different convolutional layers are simultaneously coalesced for obtaining broader information. The decoder is then altered to upsample the encoder for pixel-wise mapping from the input images to the segmented output. Experimental results reveal the efficacy of the proposed network over the state-of-the-art methods and its capability to be deployed in an average system.

Semi-supervised learning approach for Indonesian Named Entity Recognition (NER) using co-training algorithm

The problem of utilizing machine learning approach in Indonesian Named Entity Recognition (NER) system is the limited amount of labelled data for training process. However, unlike the limited availability of labelled data, unlabelled data is widely available from many sources. This enables a semi-supervised learning approach to solve this NER system problem. This research aims to design a semi-supervised learning model to solve NER system problem. A semi-supervised co-training learning is used to utilize unlabelled data in NER learning process to produce new labelled data that can be applied to enhance a new NER classi□cation system. This research uses two kinds of data, Indonesian DBPedia data as labelled data and news article text from Indonesian news sites (kompas.com, cnnindonesia.com, tempo.co, merdeka.com and viva.co.id) as unlabelled data. The pre-processing steps applied to analyze unstructured text are sentence segmentation, tokenization, stemming, and PoS Tagging. The results of this pre-process are the NER and its context used as unlabelled data for the semi-supervised co-training process. The SVM algorithm is used as a classi□cation algorithm in this process. 10 Cross Fold Validation is used as the system testing approach. Based on the result of the NER testing system, the precision is 73.6%, the recall is 80.1% and f1 mean is 76.5%.

Feature extraction for classifying lesion's shape of breast ultrasound images

The reading of ultrasound results is subjective, depending on the radiologist. Therefore, physicians need a tool as second opinion to make decision about the diagnosis of breast cancer malignancies depending on various parameters including shape parameter. To determine this parameter, a method to classify the lessions shape type of breast ultrasound image with image processing technique is proposed. In this research, two extraction feature methods, namely zernike moment and invariant moment are compared. This research also compares two methods of classifier, namely support vector machine (SVM) and multilayer perceptron (MLP). The first step is to determine region of interest (ROI) from lesion image, then the image will be filtered to reduce speckle noise. The adaptive median filter is applied to filter the input image followed by segmentation based on chan-vese active contour. Feature extraction is conducted by using Zernike moments and invariant moment, followed by classification process by using support vector machine (SVM) and multilayer perceptron (MLP). From the 45 images, the proposed method achieves 80% for classification.

Decision support system for stock trading using multiple indicators decision tree

Decision support system using decision tree classification can be used for stock trading technical analysis. Technical analysis is a stock analysis method which solely based on interpreting stocks price chart movement or trend. Historical stock prices and volume are used as variable input. The system is built based on financial market technical analysis indicators (Exponential Moving Average, Moving Average Convergence Divergence, Relative Strength Index, Money Flow Index, and parabolic Stop and Reverse). The proposed method is arrange indicators set into decision tree based on stock trading rules and it create buy, hold, and sell classes which represented decisions in trading. Decision classes then are analyzed for their profitability, geometric mean return, and cumulative wealth index. Furthermore sensitivity analysis is added into profitability analysis to obtain more positive value trading in decision making. The research purpose is to enhance decision making in technical stock trading. Compared to the single indicator decision tree multiple indicators offers 20% enhancement in decision making.