Dengan adanya pandemi COVID-19, maka protokol kesehatan seperti menjaga jarak, mencuci tangan dengan sabun secara rutin, dan menggunakan masker merupakan arahan yang diberikan oleh World Health Organization (WHO) untuk mengurangi resiko penyebaran virus COVID-19. Tetapi dengan adanya arahan tersebut, masih ditemukan orang yang tidak menggunakan masker di tempat umum. Munculnya trending Machine Learning dan Deep Learning menciptakan berbagai riset untuk menemukan metode – metode baru dan arsitektur mutakhir seperti YOLO (You Only Look Once). YOLO merupakan arsitektur detector yang diklaim sebagai “fastest deep learning object detector” yang mengorbankan akurasi dengan kecepatan. Dengan menggunakan YOLOv3, kita dapat menciptakan deteksi masker yang robust dan presisi untuk mendeteksi apakah seseorang yang tampak pada gambar / kamera bisa dikenali menggunakan masker atau tidak. Tetapi dengan tersedianya YOLOv3 yang memerlukan arsitektur komputer yang berat, maka sistem arsitektur yang lebih lama akan kesulitan menggunakan arsitektur tersebut. Maka menggunakan YOLOv3-tiny dapat menjadi solusi untuk arsitektur komputer yang lebih lama. Tentunya apabila konsekuensi YOLOv3 adalah akurasi, maka menggunakan YOLOv3-tiny tentunya akan lebih memperburuk akurasi deteksi objek.
Pattern classification is one of the relevant problems in Artificial Intelligence. Neural networks have been studied as one of the most successful methods for pattern classification. Classical perceptron can only solve linear classification problems. Morphological Neural Networks (MNN) is an alternative way to solve classification problems in the form of linear and nonlinear. Dendrite Morphological Neural Networks (DMNN) is introduced as an improved method of classical MNN. The important problem that occurs in the DMNN training algorithm is to cluster objects with hyper boxes and classify each in the corresponding class. This paper presents the proposed training algorithm using K-medoids clustering algorithm to create the hyper boxes in the dimensional space. Kmedoids is better than other clustering methods in execution time and not sensitive to outliers. The implementation of the proposed algorithm will be involved in various simulations using artificial data sets and compared with other methods to evaluate the performance of this method in future work.
In computer vision, image segmentation is a process of dividing image to get several segments of image. Image segmentation aim to divide image into simple section that meaningful and easy to analyze. Image segmentation regularly use to locate boundary of object in image, so object in image can analyzed. Object area boundary recognized from edge of object, and the edge of object in the image can be recognized from high discoloration. Superpixel is one of popular image segmentation methods. Superpixel has commonly partitioning image into simply part and reduce computation in various computer vision task. On the other side cosine similarity is a method used to analyze the similarity of two objects based on its features. This paper proposed an image segmentation using Superpixel and the sine similarity. Superpixel used to process segmentation and partitioning image into mere parts. Each segment labeled and process to compare with each neighbor using the cosine similarity. The experiment result shows that cosine similarity can be used to recognize similar segment from superpixel segmentation and make the boundary of object more significant.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.