Modified Convolutional Neural Networks (CNN) with MobileNet-v2-186 Architecture in Crescent Moon Image Detection

UNSPECIFIED Modified Convolutional Neural Networks (CNN) with MobileNet-v2-186 Architecture in Crescent Moon Image Detection.

Preview

Text habib_artikel_conference.pdf Download (1MB) | Preview

Abstract

The use of image processing in rukyatul hilal (crescent moon observation) is highly important, since the observed crescent images often suffer from quality degradation. This indicates the presence of defects or noise, resulting in excessively high contrast and image blurring. Digital image processing operations are essential for addressing challenges encountered in the observation of crescent moon imagery. This study aims to detect the appearance of crescent moon imagery in real time by developing and modifying Convolutional Neural Networks (CNN). The development and modification of the CNN model were based on a series of experiments, including the adaptation of the architecture using MobileNet-v2-186. The research dataset consisted of 27 crescent moon observation videos, comprising a total of 5,241 training frames and 1,042 testing frames. Experimental results indicated that the MobileNet-v2-186 model achieved an accuracy of 99.6835%, a precision of 82.1152%, a recall of 82.2122%, and an F1 score of 80.5725%. This development was undertaken to enhance the detection accuracy of crescent moon images, which exhibit extremely subtle visual characteristics and are difficult to identify manually.

Actions (login required)

View Item