Görüntü İşleme ve Derin Öğrenme Tabanlı Aydınlatma Şiddeti (Lüx) Kestirimi: MobileNetV2 Mimarisi ile Bir Uygulama

Résumé fourni par la source

This study presents a deep learning-based approach for estimating illuminance (Lux) from ambient photographs with high accuracy, as an alternative to physical luxmeter sensors. A unique dataset consisting of 729 ambient images at 1482x855 resolution and their corresponding lux values was used in the study. A customized cropping algorithm was developed to reduce noise (walls, ceilings, dead zones) in the images. The model architecture used the MobileNetV2 network, proven in image classification, and adapted it to the regression problem via transfer learning. After training, the model reduced the Mean Absolute Error (MAE) value to 0.78 Lux on the validation dataset. Furthermore, the model's R^2-score demonstrated high stability. The findings indicate that the developed method can precisely measure ambient illuminance using only camera images, without the need for expensive hardware.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.