An Advanced Deep Learning Approach for Multi-Object Counting in Urban Vehicular Environments

Object counting is an active research area that gained more attention in the past few years. In smart cities, vehicle counting plays a crucial role in urban planning and management of the Intelligent Transportation Systems (ITS). Several approaches have been proposed in the literature to address this problem. However, the resulting detection accuracy is still not adequate. This paper proposes an efficient approach that uses deep learning concepts and correlation filters for multi-object counting and tracking. The performance of the proposed system is evaluated using a dataset consisting of 16 videos with different features to examine the impact of object density, image quality, angle of view, and speed of motion towards system accuracy. Performance evaluation exhibits promising results in normal traffic scenarios and adverse weather conditions. Moreover, the proposed approach outperforms the performance of two recent approaches from the literature.

Deteksi Malaria Berbasis Segmentasi Warna Citra dan Pembelajaran Mesin

<p class="Abstrak">Di beberapa daerah di Indonesia, malaria masih merupakan salah satu penyakit endemik dan termasuk ke dalam kategori penyakit menular dengan vektor nyamuk <em>Anopheles</em>. Penurunan jumlah mortalitas penderita malaria ini telah menjadi program Pemerintah Indonesia dan <em>World Health Organization</em>. Salah satu hal penting yang dapat dilakukan adalah menyediakan alat diagnosis malaria yang cepat dan akurat berbantukan komputer. Oleh karena itu, pada studi ini dikembangkan sebuah metode deteksi malaria berbasis segmentasi warna citra yang dikombinasikan dengan metode pencacahan objek citra dan pembelajaran mesin berbasis <em>Convolutional Neural Network</em>. Pada studi ini, segmentasi citra dilakukan dengan menetapkan suatu nilai ambas batas tertentu (<em>thresholding</em>) pada model warna HSV. Nilai ambang batas untuk masing-masing kanal warna ditetapkan sebagai berikut: H = 100-175, S = 100-250, dan V = 60-190. Terdapat tiga skema pembelajaran mesin yang digunakan, yaitu citra asli menggunakan <em>RMSProp</em> <em>optimizer</em>, citra tersegmentasi menggunakan <em>RMSProp</em> dan <em>Adam</em> <em>optimizer</em>. Akurasi pelatihan dan validasi CNN tertinggi diperoleh dengan skema citra tersegmentasi menggunakan <em>RMSProp</em> <em>optimizer</em>, yaitu sebesar 92,77% dan 94,38%. Sementara, deteksi malaria berbasis pencacahan objek memiliki akurasi sebesar 93,78%. Meskipun deteksi malaria berbasis pencacahan objek memiliki akurasi 93,78%, tetapi sumber daya komputasi dan waktu yang diperlukan jauh lebih rendah.</p><p class="Abstrak"><strong><em>Abstract</em></strong></p><p class="Abstrak"><em>Malaria is still one of the endemic diseases in several regions of Indonesia. Reducing the malaria mortality rate has become a notable programme, not only does the Government of the Republic of Indonesia project it, but also the World Health Organization has a similar plan to tackle this disease. One of the prominent concerns to properly promote this programme is providing a rapid and accurate malaria diagnosis tool by applying the computer-aided diagnostics to minimize human errors. The aim of this study is to develop a colour microscopic image-based malaria detection using object counting and CNN-based machine learning. In this research, the HSV colour model with threshold values of H: 100-175, S: 100-250, and V: 60-190 was used to remove the image background. There are three machine learning schemes implemented in this study, i.e. original image using RMSProp optimizer, segmented image using RMSProp and Adam optimizer. The highest training and validation accuracy of CNN were obtained using a segmented image scheme by the RMSProp optimizer, 0.9277 and 0.9438. On the contrary, object-based malaria detection has an accuracy of 93.78%. Furthermore, there are several considerations to determine the malaria detection method, i.e. accuracy, computational resources, and time. Even though malaria detection using object counting has an accuracy of 93.78%, lower than the accuracy of CNN validation, the computational resources and time required are much lower and faster. Therefore, this detection method is suitable for smartphone-based devices with low-middle end specifications.</em></p>