RWRangga Winantyo

Rangga Winantyo

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Rangga Winantyo has a total of 8 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Real-Time Meat Authentication Using Deep Convolutional Neural Network with ResNet-50 Architecture

    by Adhi Kusnadi, Rangga Winantyo, Fenina Adline Twince Tobing, Muhammad Tanveer, Ghulam Jilani Waqas

    Background: Many issues arise from meat adulteration with regard to food safety and halal certification. Because pork is so visually similar to beef, removing or identifying either type of product becomes almost impossible without some sort of testing. Unfortunately, for many situations, traditional methods for testing authenticity are typically lengthy, costly, and impractical. Objective: This study utilizes computer vision to authenticate beef, pork, and mixtures of both using a deep learning approach. The method used consisted of preprocessing the photographs and performing various forms of data augmentation to increase the dataset’s representative and generalizing capabilities. Methods: We achieved our highest classification accuracy through transfer learning using ResNet-50 as the primary model used for classifying the different types of meats covered by the study, providing empirical metrics and K-fold cross-validation to support reliable classification results. A comparative evaluation using MobileNetV2, EfficientNetB0, and InceptionV3 was also conducted to validate the effectiveness of the proposed ResNet 50 model. Results: The experiment showed that the proposed ResNet-50 model could classify images into their respective classes with a high level of accuracy (98.33%) and greater than 97% for both precision and recall across all classes, in addition to an average inference time of 0.124 s per image. Furthermore, the 5-fold cross-validation yielded an average classification accuracy of 99.33% ± 0.37%, indicating that the performance was stable across multiple data partitions. The low inference time demonstrates the feasibility of the proposed model for real-time meat authentication applications on mobile devices. Conclusion: Overall, the findings indicate that the proposed approach is effective for meat product verification and that computer vision solutions can support the development of practical food safety and halal verification applications. Keywords: Meat authentication, Deep learning, Convolutional Neural Network, ResNet-50, Real-time classification, Food safety

  • Open Access

    PCA effect on the 3D face recognition system speed

    by A Kusnadi, Wella Wella, R Winantyo, I Z Pane

    In this paper, a system of three-dimensional (3D) face recognition is not done through 3D face reconstruction method but directly uses the data retrieved from Kinect Xbox camera system. From a previous study, there exists a possibility to increase the speed and accuracy of the system. In order to accelerate the recognition speed, a single step in the said study is eliminated, which is the reconstruction of 3D face data. The algorithms used in this research are Backpropagation Neural Network and PCA. Testing is done in two ways. The first test uses a combination of Backpropagation and PCA, while the second test uses Backpropagation only. PCA has the function of simplifying the amount of data, thereby reducing the amount of computing that can increase speed system. The results show that the system that uses a combination of Backpropagation and PCA achieved an increase in speed up to 34.2 times, but a reduced accuracy of 8.5%.


  • Removing DCT High Frequency on Feature Detector Repeatability Quality

    by Adhi Kusnadi, Wella Wella, Rangga Winantyo, Ivransa Zuhdi Pane

    This study evaluates five feature detectors popular used in 3-dimensional face recognition systems, where face images quality is improved by removing high frequencies at the Discrete Cosine Transform coefficient domain, because at this frequency contained noise. The evaluation is important to be done because the feature detector functions to look for keypoints on the face image needed in 3-dimensional face reconstruction from 2 dimensional face images, and th e accuracy of 3d face recognition depends on the success of the reconstruction. Several researches had been undertaken in 3D face recognition, among others by implementing feature detector to determine the same key point on two face images 2 dimensional at different angles to simulate epipolar geometry. However, the implementation of feature detector does not result in a maximum value of reliability, and thus improvements should be made, it could be on feature extraction and on feature selection stages. Discrete Cosine Transform algorithm is used as a feature extractor, which produced a coefficient consisting of low frequency, middle frequency and high frequency bands. A test of five feature extractors was carried out by removing the high frequency band from the coefficient as these frequencies. The face images were taken from ORL database and Head Pose Image Database, then data before and after the high frequency bands were removed are compared. SURF attained the highest F Score for both conditions, and on average, the value F score increases from the condition before the high frequency is removed and after it is removed from the DCT coefficient

  • Evaluation of Feature Detectors on Repeatability Quality of Facial Keypoints In Triangulation Method

    by Adhi Kusnadi, Wella Wella, Rangga Winantyo, Ivransa Zuhdi Pane

    This study derived from a research focusing on 3D face recognition using ToF camera. But the system can't be used outdoors, because of a backlight. To solve this problem, a commercial digital single-lens reflex (DSLR) camera will be used. It can be approached y solving the stereo-view reconstruction problem for each pair of consecutive images. To reconstruct an object, projection matrix estimation from 2D point correspondences will be needed. The accuracy of 3D reconstruction is highly dependent on the corresponding points of 2D data projections from images to other images. In this research, The detectors are Harris-Stephens, SURF, FAST, Minimum Eigenvalue, and BRISK have been tested and analyzed through black box test. To evaluate feature detectors performance, the repeatability score for a given pair of images is computed. To do that it can use recall and precision. The best detector is the Harris Stephens detector because it has the best F-measure values of 0.46.