A DEEP LEARNING MODEL FOR DETECTION OF MULTI FOG TYPES IMAGES

• Main Author: ZAINAB HUSSEIN ARIF
• Format: text::Thesis
• Language: English
• Published: 2023

The fog has different characteristics and effects within every single environment. Detecting fog in the image considered as a challenge and giving the type of fog has a substantial enlightening effect on image defogging. Foggy scenes have different types such as scenes based on fog density level and scenes based on fog type. Machine learning in general and especially deep learning techniques have significant contribution in the detection of foggy scenes. Nevertheless, most of the existing detection models are based on traditional machine learning models which are not efficiently dealing with huge volume of input data and depend on complex feature extraction methods comparing with deep learning models, however, only a few studies have adopted deep learning models for foggy image detection task. Furthermore, most of existing machine learning detection models are based on the fog density level scenes, while more complex foggy scenes should be considered. However, to the best of our knowledge, detection model based on multi-fog types scenes have not been explicitly addressed by literature studies yet. Therefore, this study aims to propose a deep learning model for detection of multi-fog types of images. Moreover, due to the lack of publicly available dataset for inhomogeneous, homogenous, dark and sky foggy scene, a dataset for multi-fog types scenes is presented in this study. Experiments were conducted in three stages. First, the data collection phase based on eight resources to obtain the multi-fog types of scenes dataset. Second, a classification experiment is conducted based on ResNet-50 deep learning model to obtain detection results. Third, the evaluation phase where the performance of ResNet-50 detection model has compared against three different models. Experimental results show that the proposed model can accurately detect different foggy images with 96% detection rate, which has specific theoretical and practical significance. Our proposed model is suitable as a pre-processing step and might be considered in different real-time applications.